A Short History of Structural Linguistics

This book is a concise history of structural linguistics, charting its

development from the 1870s to the present day. It explains what
structuralism was and why its ideas are still central today. For
structuralists a language is a self-contained and tightly organised
system whose history is of changes from one state of the system to
another. This idea has its origin in the nineteenth century and was
developed in the twentieth by Saussure and his followers, including
the school of Bloomfield in the United States. Through the work
of Chomsky, especially, it is still very influential. Peter Matthews
examines the beginnings of structuralism and analyses the vital role
played in it by the study of sound systems and the problems of how
systems change. He discusses theories of the overall structure of
a language, the Chomskyan revolution in the 1950s, and the
structuralist theories of meaning.
is Professor of Linguistics at the University of
Cambridge. He is the author of Syntax (1981), Morphology (2nd edn,
1991) and Grammatical Theory in the United States: From Bloomfield
to Chomsky (1993), also published by Cambridge University Press.

HH

A Short History of
Structural Linguistics
Peter Matthews
Professor of Linguistics, University of Cambridge

PUBLISHED BY CAMBRIDGE UNIVERSITY PRESS (VIRTUAL PUBLISHING)

FOR AND ON BEHALF OF THE PRESS SYNDICATE OF THE UNIVERSITY OF
CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge CB2 IRP
40 West 20th Street, New York, NY 10011-4211, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
http://www.cambridge.org
Peter Matthews 2001
This edition Peter Matthews 2003
First published in printed format 2001

A catalogue record for the original printed book is available

from the British Library and from the Library of Congress
Original ISBN 0 521 62367 7 hardback
Original ISBN 0 521 62568 8 paperback

ISBN 0 511 01151 2 virtual (eBooks.com Edition)

In memoriam
R. H. R.
19212000

HH

Contents

Preface
1

Introduction

Languages
2.1 Linguistics as the study of language systems
2.2 Languages as sets of utterances
2.3 The autonomy of linguistics

Sound systems
3.1 The prehistory of the phoneme
3.2 Phonology
3.3 Structuralism

Diachrony
4.1 Diachronic phonology
4.2 System and norm
4.3 Universals

5
10
20
25

31
32
40
48

52
55
61
69

74

5.1 Expression and content

5.2 Phonology and grammar
5.3 Deep structure and surface structure

75
81
88

Internalised language

Structural semantics
7.1 Meanings as invariants
7.2 Semantic fields
7.3 Semantic interpretations

The architecture of a language system

6.1 Generative grammars

6.2 Knowing a language
6.3 Universal Grammar and diachrony

page ix

Structuralism in 2000

References
Index

96
97
103
113

118
119
126
133

142
154
160
vii

HH

Preface

The last chapter of this book was finished on the day when Bobby Robins, whose A Short History of Linguistics has been admired for more than
thirty years, was found dead. It is with sadness and affection that I
dedicate it, with its presumptuously similar title, to his memory.
I was initially not at all sure how this history should be written: in
particular, how selective and, in consequence, how long it should be. For
advice at that stage I am especially grateful to Jeremy Mynott and, in a
sense that they will understand, to my fellow editors of the Cambridge
Textbooks in Linguistics. Conversations with Kasia Jaszczolt have since
helped, at various times, to clarify my thinking.
I am also grateful to Andrew Winnard for waiting patiently for the
book to be written.

ix

HH

Introduction

Introduction

What is structural linguistics? Do most linguists still accept its principles? Or are they now believed in only by old men, clinging to the ideas
that were exciting in their youth? Who, among the scholars who have
written on language in the twentieth century, was or is a structuralist?
Who, by implication, would that exclude?
It may seem, at the outset, that the first of these questions should be
fundamental. We must begin by asking what, in general, we mean by
structuralism. There are or have been structuralists in, for example,
anthropology; also in other disciplines besides linguistics, such as literary criticism and psychology. What unites them, and distinguishes them
from other theorists or practitioners in their fields? In answering this
question we will identify a set of general principles that structuralists
subscribe to; and, when we have done that, we will be able to ask how
they apply to the study of language. From that we will deduce the tenets
that a structural linguist should hold; we can then see who does or, once
upon a time, did hold them. But an inquiry in this form will lead us only
into doubt and confusion. For different authorities have defined structuralism, both in general and in specific application to linguistics, in what are
at first sight very different ways. There are also linguists who are structuralists by many of the definitions that have been proposed, but who
would themselves most vigorously deny that they are anything of the kind.
Let us look, for a start, at the definitions to be found in general dictionaries. For structuralism in general they will often distinguish at least
two different senses. Thus, in the one-volume Collins (1994 edn; originally
Hanks, 1979), an approach to linguistics (sense 2) has one definition and
an approach to anthropology and to other social sciences and to literature (sense 1) has another; and, for a reader who does not know the
problems with which the editor had to deal, it is not obvious how they
are connected. In anthropology or literature, structuralism is an approach
that interprets and analyses its material in terms of oppositions, contrasts, and hierarchical structures, especially as they might reflect universal mental characteristics or organising principles. Compare, we are
1

Introduction

told, functionalism. In linguistics, it is an approach that analyses and

describes the structure of language, as distinguished from its comparative
and historical aspects. The next entry defines structural linguistics in
terms that are in part different and in part supply more detail. It is, first
of all, a descriptive approach to a synchronic or diachronic analysis of
language. But a diachronic analysis is precisely one that deals with
historical and, where they are a source for our knowledge of the history,
comparative aspects. This analysis, to continue, is on the basis of its
structure as reflected by irreducible units of phonological, morphological,
and semantic features. This seems to imply that the units that structural
linguists establish are necessarily of these three kinds.
The New Shorter Oxford English Dictionary (Brown, 1993) distinguishes
two main senses of structuralism, one in early twentieth-century psychology (compare Collins under structural psychology), the other covering
all other disciplines, but with specific subsenses (2 (a), 2 (b) and 2 (c) )
in linguistics, in anthropology and sociology, and as a method of critical
textual analysis. In sense 2 in general, structuralism is any theory or
method which deals with the structures of and interrelations among
the elements of a system, regarding these as more significant than the
elements themselves. It is also, by a second or subsidiary definition, any
theory concerned with analysing the surface structures of a system in
terms of its underlying structure. So, specifically in linguistics (sense 2
(a) ), it is any theory in which language is viewed as a system of interrelated units at various levels; especially, the definition adds, after the
work of Ferdinand de Saussure. There is nothing in this entry about
synchrony or diachrony. But under structural (special collocations), structural linguistics is defined, in terms which recall the Collins definition
under structuralism, as the branch of linguistics that deals with language as a system of interrelated elements without reference to their
historical development. Thus, by implication, structuralism in linguistics
is again not diachronic. One is also left wondering about the reference to
surface and underlying structure. The term underlying is picked up, in
the subdefinition for anthropology, with reference to the theories of Claude
Lvi-Strauss (concerned with the network of communication and thought
underlying all human social behaviour); but not specifically for linguistics.
However, in the Supplement to the main Oxford English Dictionary, which
is the immediate source of these definitions, the term structural is also
said to mean, under sense 5a, relating to or connected with the deep
structures that are considered to generate surface structures.
These are good dictionaries, and I am not out to criticise them. I can
hardly claim that the entry in my own concise dictionary of linguistics
(Matthews, 1997: 356f.) is more definitive. For the root of our difficulty is

Introduction

that linguists themselves do not apply these terms consistently. In a leading

survey of the subject, Giulio Lepschy suggests that structural linguistics
has at least three possible senses (Lepschy, 1982 [1970]: 35f.). But of these
one, as he in effect remarks, is vacuous. Another applies so narrowly that
most of what has generally been perceived as structuralism does not fall
within it. The third remains, as a definition, tantalisingly general.
In the widest sense, with which Lepschy begins, every reflection on
language has always been structural. In any grammar, for example, units
are identified; units of any one kind are related to others of their own or
of another kind; and through these relations, which will be in part hierarchical, successively larger structures are quite clearly formed. In that
sense, any synchronic or diachronic analysis of language (Collins) cannot
but be structural. Hence, for Lepschys and our purposes, this first use
of the term is scarcely revealing (1982: 36).
Lepschys narrowest sense dates from the 1960s, when the American
linguist Noam Chomsky was attacking what he called the taxonomic
methods of his predecessors. The charge was levelled against a specific
school in the United States, who were also accused at the time, in apparent variance with a hint in one of our dictionaries, of a concern with no
more than the surface structures of language, to the exclusion of its
deep structures. For Chomsky and his followers, structuralists were
above all members of that school. Hence, in some accounts, like that of
David Crystal in The Cambridge Encyclopaedia of Language, the term
structuralist is used only of them and structural(ist) linguistics only of
a limitation of the subject in a way that they alone proposed (Crystal,
1997 [1987]: 412; glossary, 438).
The middle sense refers, in Lepschys words, to those trends of linguistic thought in this [the twentieth] century which deliberately tried to
gain an insight into the systematic and structural character of language.
This is indeed more widely accepted (36) than the largest sense with
which he began. But Lepschys wording again leaves one wondering whether structuralism can be defined precisely. For no one will deny that language has a systematic and structural character; and, as we move into a
new century, many scholars are still seeking to understand it. Yet Lepschy
refers to trends that tried, in the past tense, to do so. What is it that
those trends specifically, which are by implication characteristic of the
twentieth century, had in common? What were the particular insights, or
the particular ways of trying to gain an insight, that lead us to distinguish
them from other trends that are not structural?
Lepschys Survey of Structural Linguistics is the best book of its kind,
and I am not seeking to pick holes in it. For what this makes clear is that
structuralism has to be defined, in part, historically. The term structural

Introduction

linguistics dates, as we will see, from the late 1930s, and referred to an
intellectual movement that was by then well established. But it had no
single leader, and no wholly uniform set of principles. In the view of most
continental Europeans, it had been founded by Ferdinand de Saussure,
whose lectures on general linguistics (Cours de linguistique gnrale) had
been reconstructed and published after his death in 1913. Hence the specific reference to him in The New Shorter Oxford English Dictionary. But
structural and structuralisme were not terms that Saussure had used.
Therefore he had not laid down the principles, by name, of structuralism, and the ideas that he had expounded were already being developed,
by different scholars, all of whom could reasonably claim to be his followers, in varying directions. In the United States, by contrast, linguists
who were young at the end of the 1930s were influenced above all by
the American scholar Leonard Bloomfield, whose great book Language
had appeared in the first half of the decade. But he did not talk of
structuralism either. Nor did the theory that he propounded agree
entirely with Saussures. By the time the movement had a name the trends
(plural) to which Lepschy refers could already be distinguished.
But, as a broad movement, it quite clearly existed. Structuralists in general, of whatever more precise persuasion, came to be lumped together by
their critics; and, among the structuralists themselves, there was a sense of
unity. A political party, if we may take one obvious parallel, includes many
shades of opinion. It would again be hard to say exactly what set of beliefs
its members all have in common, from one time to another or even at any
one time. But the trends within it form a network of shared interests and
shared inspirations, in which all who belong to it have some place. With
intellectual movements, such as structuralism, it is often much the same.
Or should we say, in this case, that it was the same? Lepschy used,
once more, the past tense; and it is now more than thirty years since he
was writing. But on the next page he speaks of Chomskys theories,
which had by then come to dominate the subject, as from his perspective an heir to . . . structural linguistics and one of its most interesting
developments. There is no doubt that, by the end of the 1960s, the sense
of party unity had been lost, at least between Chomsky and the older
generation in the United States. But the implication is that structuralism,
in a broad sense, passed into a new phase. Has there, since then, been a
real break? Or is the thinking of most scholars now, about what Lepschy
called the systematic and structural character of language, still continuous with the tradition that was dominant earlier?
I will return to these questions in the final chapter. But first we have
more than a hundred years of history, and the thought of some of the
best minds that have studied language, to work through.

Linguistics as the study of language systems

Languages

Linguistics is said in dictionaries to be the branch of knowledge that

deals with language (New Shorter Oxford English Dictionary) or the
scientific study of language (Collins). But for structuralists it has been
much more the study of, in the plural, languages. This was true at the
outset, for Saussure, and is still true for many as we enter the twenty-first
century. What then constitutes a language? It is easy to give examples:
English is one, Japanese another, and so on. But what, in general, are they?
Let us look again at dictionaries. For the first editor of The Oxford
English Dictionary (Murray et al., 1933 [1884 1928] ), the earliest sense of
language (1) was that of the whole body of words and of methods of
combination of words used by a nation, people, or race; alternatively, a
tongue. The dictionary itself was thus an account of the whole body of
words that constitute the lexicon of English. The second definition (2)
adds a generalized sense: words and the methods of combining them
for the expression of thought. But where Murray saw a body, The New
Shorter Oxford Dictionary speaks of a system. Language is a system
of human communication using words . . . and particular ways of combining them; it is any such system, the definition adds, employed by a
community, a nation, etc. (1a). In the Collins dictionary, it is a system
for the expression of thoughts, feelings, etc. by the use of spoken sounds
or conventional symbols (1); also in general (2) the faculty for the use
of such systems. These accounts have much in common. In particular, a
specific language is related, either by definition or by historical association, to a nation or other community. But a system is potentially
more than a body. A body, in the sense that Murray must have had in
mind, can be described by an inventory. A dictionary is thus an inventory
of words, arranged for convenience in alphabetic order. A grammar is
in turn an inventory of methods of combining words, arranged perhaps
by classes to which combinations can be assigned. But a system is not
simply a collection of individual components. Suppose that, from an
inventory, we omit one item: say, from the inventory of words in English,
we omit the word we. The remainder of the inventory is unchanged. But
5

Languages

if a language is a system then, as part of that system, we is related to

other words: most obviously to I, you, us and others that are traditionally
called pronouns. If we is omitted, the relations that the other pronouns
enter into must in turn change.
It is with this basic insight that, in the closing decades of the nineteenth
century, structuralism began. It did not develop fully until later, according to most commentators with the publication of Saussures Course in
General Linguistics. But by then this insight had already informed the
study of sound systems, as we will see in the next chapter (3.1). We can
also find at least one earlier and independent programmatic statement, in
an introduction to linguistics of the early 1890s by the German Orientalist Georg von der Gabelentz. Every language, he writes, is a system
all of whose parts interrelate and interact organically (Jede Sprache
ist ein System, dessen smmtliche Theile organisch zusammenhngen und
zusammenwirken). Thus, in our example, we relates to and interacts not
just with I and you but, directly or indirectly, with all other elements of
the wider system of which pronouns are part. One has the impression,
Gabelentz continues, that none of these parts could be missing or be
different, without alteration to the whole (Man ahnt, keiner dieser Theile
drfte fehlen oder anders sein, ohne dass das Ganze verndert wrde)
(Gabelentz, 1901 [1891]: 481). Thus, if there were no pronoun we, the
repercussions would extend throughout English generally. In a famous
Saussurean formula, a language is a system in which everything holds
together (un systme o tout se tient). Change again one element, and
the system is different.
The origins of this formula have been explored by Konrad Koerner, in
an essay dealing with the connections between Saussure and the French
Indo-Europeanist Antoine Meillet. It is not, on paper at least, Saussures
own. But Meillet was a young man in the 1880s; he had heard lectures
by Saussure in Paris; and by 1893 he was saying already that the units of
sound in each form of speech (les divers lments phontiques de chaque
idiome) form such a system. The point can be appreciated, he remarks,
by anyone who has tried to learn the pronunciation of another language.
But it must also apply to children learning their first language: A child,
in learning to speak, assimilates not an isolated articulation, but the
whole of the system (Or lenfant, en apprenant parler, sassimile non
une articulation isole, mais lensemble du systme). The passage is cited
by Koerner (1989 [1987]: 405), and Saussure was not mentioned. But the
formula fits so beautifully with the ideas developed in Saussures Cours
de linguistique gnrale that, as Koerner points out, it was later cited
as if it were his. For the Russian linguist N. S. Trubetzkoy, writing in the
early 1930s, this conception of a language was one of the basic principles

Linguistics as the study of language systems

that Saussure had proclaimed (Trubetzkoy, 1933: 241; also 243, with the
formula cited as if it were a quotation).
Like Meillet, Saussure was a student of Indo-European, the vast family
that links most languages in Europe with most of those from Persia to
Southern India. It was established early in the nineteenth century that
these had developed from a common prehistoric language; but it was not
until the 1870s, when Saussure was a student in Leipzig, that, in Leipzig
especially, the structure of that language was first satisfactorily reconstructed. It was not that of any ancient language historically attested: not
that of Latin, nor of Greek, nor even, as had still been assumed in some
important respects in the 1860s, of the ancient Indian language Sanskrit.
Nor was its reconstruction simply a matter of comparing individual units.
It was precisely the structure that was recovered. Saussures first book
was written in Leipzig, and was itself a striking contribution to this enterprise. It is therefore worth our while to glance at some of the details.
Let us begin with a specific problem that Saussure could take as solved.
In Ancient Greek, for example, the accusative singular usually ends in -n:
hod-n road or oikA-n house. Compare -m in Latin (dominu-m master
or puella-m girl) or in Sanskrit (dev-m god). But in Greek it could also
be -a: thus in the words for mother and father (mBtr-a, patr-a). Is this
simply an irregularity, by which some nouns in Greek decline aberrantly?
At first sight it is: in Latin, for example, the corresponding forms again
end in -m (matre-m, patre-m). But let us suppose, as a hypothesis, that in
the prehistoric language the ending was throughout *-m. It is marked
with an asterisk, to show that this is a reconstruction and not, for instance,
the historical -m of Latin. But phonetically the consonant had, we can
assume, a nasal articulation, which is preserved in both the -m of Latin
dominu-m and the -n of Greek hod-n. Let us also suppose, as a further
hypothesis, that the phonetic element *m was neither simply a consonant
nor simply a vowel. Instead it was one that could, in general, either
accompany a vowel to form a syllable (consonant + vowel + m, m +
vowel, and so on) or, itself, have the position of a vowel within one
(consonant + m, or consonant + m + consonant). In that respect it is like,
for example, the n in spoken English, which forms a syllable with t,
again with no vowel sound, in a word like, in phonetic spelling, [bktnh]
(button) or [bktnhhivl] (buttonhole). The apparent irregularity will then
make perfect sense. In the form that prehistorically underlay, for example, hodn the ending *-m came after a vowel and developed in Greek
into -n. In the form that underlay, for example, mBtra it came after a
consonant (consonant + m). In that context it became, instead, -a.
For an account of this period I must defer to the masterly history of
nineteenth-century linguistics by Anna Morpurgo Davies (1998 [1994]:

Languages

Ch. 9): the solution outlined is one facet of a wider hypothesis developed
by Karl Brugmann and Hermann Osthoff (Morpurgo Davies, 1998: 242f.).
But it is plain already that the argument does not affect a single unit.
That the prehistoric language had a sound *m was not new: it was obvious enough, at the beginning of a syllable, from sets of words like those
for mother, honey (Greek mli, Latin mel), and so on. What matters
are the relations in which it is claimed to have stood to other units. They
are wider than those borne by any unit, such as m in Latin, that has
hypothetically developed from it. But, given its role as reconstructed, it
was possible to explain, by different historical developments in different
languages and in different positions in the syllable, what would otherwise
remain puzzling.
The next step, or what with hindsight seems most logically to have
been the next step, was to posit units in the prehistoric language that are
not directly attested. In Greek, by the hypothesis we have outlined, *m
changed, in the position of a vowel, to a. To be precise, it merged with it;
so that, from the direct evidence of mBtra and other such forms, we
cannot know that anything other than an a had ever been there. But if
a unit can lose its identity in one position, it can lose it in all. This can
happen in just one member of the family; and, in that case, evidence for
it will emerge when forms in which it had been present are compared
with corresponding forms in other languages. But it could also happen,
by a series of connected or independent changes, in all members known
to us. Is it possible, in that case, that it might still be reconstructed?
It was Saussure who first showed how it might. In the case of *m the
evidence we cited is of an irregularity: between, in Greek itself, the -a
of accusative singulars such as mBtr-a and the -n of, for example, hodn; and, across languages, between the -a of mBtra and the -m of, for
example, Latin matrem. But by essentially similar reasoning it is possible
to explain a whole sheaf of irregularities, many at first sight unconnected,
by positing what specialists in Indo-European call laryngal elements. In
Greek, for example, the verb for to put has a long B in some forms and
a short e in others. Compare t-thB-mi I put (thB-) with adjectival the-ts
placed (the-). The historical explanation rests in part on the hypothesis
that, in the prehistoric language, there were other elements that could
appear in either position in a syllable. Some, like *m, were directly
attested: for example, in Greek lep-D I leave behind the i derives hypothetically from a *y which follows a vowel (*leyp-), while in -lip-on I left
behind it derives from the same unit *y, but in the position of a vowel
(*lyp-). Of others there was, in the Indo-European languages as they were
known in the 1870s, no direct trace. But suppose that, in the prehistoric
language, the form for to put had such an element. We have no evidence

Linguistics as the study of language systems

of its phonetic character; all we are saying is that it fitted into a certain
system of relations. So, in the form underlying Greek thB- this element
(call it, for the sake of a symbol, *H) came after a vowel: *theH-. By a
subsequent sound change, *eH became, in Greek, B. In the case of the-,
the underlying form was hypothetically *thH-; then, in the position of
a vowel, *H became e. The variation between *theH-, changing to thB-,
and *thH-, changing to the-, is thus, so far as its form is concerned, precisely like that of, for example, *leyp-, changing to leip-, and *lyp-, changing to lip-.
Decades later, remains of the Hittite language were discovered at an
archaeological site in Turkey; it was shown to be Indo-European, and
in it, for the first time, there was direct evidence that laryngals such as
*H had existed. But the seeds of most of what we have said in the last
paragraph were sown by Saussure at the end of the 1870s, when such
elements could be established only as terms in a prehistoric system. They
could not, like *m, be given a phonetic value. The hypothesis was simply
that each was a unit and bore certain relations, in the structure of a
syllable, to other units.
Saussure was twenty-one when this work appeared (Saussure, 1879).
Unfortunately, he published very little after it, and from the 1890s, when
he returned from Paris to a chair in his native Geneva, almost nothing.
It is therefore unsafe to speculate too much about the route that might
have led him from this early work on Indo-European to the ideas for
which he is later famous. But what was reconstructed was a system of
relations among units. Each of the historical languages had a different
system. Therefore what changed, in the development of Greek, etc., from
the prehistoric language, was in each case more than just an inventory
of units. Now the historical languages were known to us through texts
associated with specific communities. They thus had an identity in time
and place, independent of the system that their units formed. Of the
prehistoric language we otherwise know nothing. It is constituted solely
by the system that we are able to reconstruct.
It is unsafe, I repeat, to speculate about a train of thought that we
cannot document. But the view that Saussure in the end reached was not
simply that a language has, or that its units form, a system. As in the
passage cited earlier from Georg von der Gabelentz, it quite literally is a
system: Jede Sprache ist ein System. Hence, at a long remove, the dictionary definitions cited at the beginning of this chapter. Hence also two
immediate conclusions, both of which Saussure, in particular, drew.
First, if languages are systems they are, from an external viewpoint,
closed. Each will have a determinate set of basic units, and a determinate set of relations among them, and will be distinguished sharply both

10

Languages

from other languages and from anything that lies outside such systems.
Therefore the study of each individual language is separate from that
of any other individual language; and within linguistics, if conceived
more widely as the investigation of all aspects of human speech, that
of individual languages must form a distinct science. In Saussures terms
this is a linguistique de la langue (a linguistics of languages), which is
autonomous and whose object is limited to what we may call language
systems.
Secondly, if everything in a system holds together, any change which
affects it will result in a new and different system. In the prehistoric
Indo-European language *m entered, hypothetically, into one set of relations. In the development of Greek it changed, in one position in the
structure of the syllable, to a. This may not have affected the inventory
of elements; but, in consequence of this one change, m in Greek now
entered into a new set of relations, the roles that a had in the structure
of the language were different, the accusatives of distinct declensions
of noun diverged, and so on. The study of systems must, accordingly,
be separated strictly from that of historical relations between systems.
As historians, we can describe the changes that relate, for example, the
Indo-European system to the Greek, or, in historical times, the Greek
system as it was in fifth-century Athens to that of Modern Greek as it
is spoken now. In Saussures terms, that is to practise diachronic linguistics, the study of languages on the time dimension. But, to be able
to carry out such studies, we must first have established the systems that
we are relating. Each system, as we have said, is different. Therefore, in
investigating, for example, the system of Modern Greek, we are not concerned in any way with that of Greek in ancient Athens, or of Greek in
any intervening period. We are concerned just with the system that exists
now. We are thus practising what Saussure called synchronic linguistics:
a pure linguistics of the language system, to which the dimension of time
and history is irrelevant.
It is now time to look in greater detail at what Saussures Cours said.
2.1

Linguistics as the study of language systems

It must be stressed at once that the book is not, in any strict sense, by
him. He gave three series of lectures on general linguistics between 1906
and 1911; but, as his literary executors were to discover, he did not keep
notes (Saussure, 1972 [1916]: 7). What we have is therefore a recration
or reconstitution (9) using all the materials available, but, in particular,
the notes of students who had followed the third course. At some points
it is based on no tangible source.

Linguistics as the study of language systems

11

If one thing ought to drive the perfectionists among us into publication,

it is the thought that when we die our lectures might be immortalised,
however objectively and conscientiously, in this way. In Saussures case, it
has given work to a small army of interpreters. Anyone who tries to
set out his ideas is faced with a text compiled and in part composed by
others, with skeletal references to the tradition in which his thought
was formed, and understood since the 1920s in every possible way. The
earliest interpretations are those of Bloomfield, Trubetzkoy, and other
structuralists in the years between the two World Wars. These are important as they are themselves part of our history. But note interpretations
(plural): there was no single reading even at that stage. Since 1960 potted
accounts have passed into textbooks, often reflecting in part the work
of later followers. But new exegesis still appears, challenging what Paul
Thibault has recently called the conventional wisdom (Thibault, 1997:
xvii). In the interval, the notes on which the Cours is based have also
been published. These were not, of course, available when it had its
greatest influence.
It is hard to read the text without this overlay of commentaries in
mind. But in the context I have outlined three notions stand out.
The first is that of linguistics as a science of language systems. If we try
to study all the phenomena of language (langage), we will open the door
to other sciences (psychology, [physical] anthropology, normative grammar, philology, etc.) that could claim it for their domain (Saussure, 1972
[1916]: 24f.). There was as Saussure saw it only one solution. We must
engage in the first instance with the individual language (la langue) and
take that as the point of reference for all other relevant phenomena
(il faut se placer de prime abord sur le terrain de la langue et la prendre
pour norme de toutes les autres manifestations du langage). Here and
elsewhere, there is an important distinction between la langue (the language or the language system) and langage (in English, language
with no article) as the wider phenomenon to which la langue is central.
Langage remains, by implication, the concern of various disciplines. But
the integral and concrete object of study in linguistics (lobjet la fois
intgral et concret de la linguistique, 23) will be la langue.
A distinction is then drawn, within langage, between la langue and
parole (speech) as the individual act of communication. When people
talk to one another a circuit is established in which their brains are
linked. Part of the link is physical (movement of sound waves); other
parts are physiological (activation of the vocal organs and the organs of
hearing); still others are mental (psychique) and involve a coupling of
acoustic images and concepts. Thus in the mind of a speaker a given
concept triggers (dclanche) an acoustic image, and in the mind of a

12

Languages

hearer the acoustic image of what is heard is associated, in its turn, with
a concept (28). But, for these concepts to match, the speaker and hearer
must have something in common. This is of its nature a social phenomenon (fait social). The physical and physiological parts of the circuit
are irrelevant. Even the mental part is not entirely relevant, since its
executive aspect (glossed parole) remains individual and not social. But
if we could encompass the totality of the verbal images stored up in all
individuals (la somme des images verbales emmagasines chez tous les
individus), we would arrive at the social bond that forms the language
(le lien social qui constitue la langue, 30). This is a stock or store (trsor)
that is built up by the experience of speech (parole) in people who
belong to the same community. It is a grammatical system that exists
potentially in each brain, or more precisely, since it is not complete in any
individual person, in the brains of the entire group.
For a psychologist of Saussures day, let alone ours, this might not
bear too close an examination. But what matters is the distinction, within
the phenomena of langage, between the language system as a social
fact (fait social), or social product (produit social) of the functioning of our intellectual faculties (30), and speech (parole) as an act that
is individual and contingent. The term fait social had been used, in
Saussures formative period, by the French sociologist mile Durkheim,
again in identifying an objective reality whose study was specific to sociology as opposed to other sciences. Take, for a simple illustration, what
we might call a system of table manners. In any society one learns how to
eat as that society expects: in Europe to hold knives, forks and spoons in
certain ways, to put them on the plate in a certain way when one has
finished, to keep ones hands at other times above or beneath the table,
and so on. The system is not that of one individual only: its existence is
in the society, in the community of people who, when on any particular occasion they are sitting at table, conform to it. But, to the point
at which their behaviour will be accepted in the society, all individuals
must individually learn it. Although this is not Saussures analogy, what
he was teaching was that, though a language is enormously more complex, its nature is essentially like that. Each member of a community has
learned it, again to the point at which they communicate as other members.
But it exists over and above the individual act of communication and the
individual communicator.
Since la langue is our first object of study, Saussure distinguished a
linguistics of the language system (une linguistique de la langue) from
a linguistics of speech (de la parole) which is secondary to it. The latter
has both a mental and a physical (that is, physiological and acoustic)
side. But the former is exclusively mental (uniquement psychique, 37).

Linguistics as the study of language systems

13

Its object of study exists in the community as a whole, in the form of a

totality of imprints registered in each brain (La langue existe dans la
collectivit sous la forme dune somme dempreintes dposes dans chaque
cerveau, 38). The study of this is linguistics in the strict sense (la
linguistique proprement dite, 38f.).
Our second notion concerns the division between synchronic linguistics and diachronic linguistics. Both belong to Saussures linguistics
of the language system. But the former is abstracted, as we have seen,
from history. It is therefore a study of the relations among elements that
form a system (des rapports . . . reliants des termes coexistants et formant
systme) as they are perceived by a single collective consciousness (tels
quils sont aperus par la mme conscience collective, 140). So, to resume
our analogy, a synchronic study of table manners would deal with those
that are general in a specific community, within a period in which they
can be treated as unchanging. A diachronic study might instead compare,
for example, the handling of knives at table in early mediaeval Europe
with their handling in later periods. It would thus deal with the relations
among successive elements that are not perceived by the same collective
consciousness (les rapports reliant des termes successifs non aperus par
une mme conscience collective). Such elements replace one another
without themselves forming a system (se substituent les uns aux autres
sans former systme entre eux, 140).
The terms synchronic and diachronic are Saussures; otherwise this
may not seem a striking insight. But for Saussure these were wholly
separate approaches (deux routes absolument divergentes), in a way
that very few linguists seemed to him to realise (114). Accordingly great
care is taken to explain the differences between them: in the abstract,
through a glance at the history of linguistics, through examples, and
through his own analogies. Let us start with one of Saussures illustrations (120f.). It was and is a stock example, and what the Cours says is, at
a certain level, what any lecturer on the history of language might have
been expected to say. But, in granting that, we can appreciate the shift of
emphasis that his followers were to seize on.
In English, then, a plural noun is generally formed by adding -(e)s:
trees from tree, grasses from grass, and so on. But in just a few the vowel
of the singular is changed: feet from foot, teeth from tooth, geese from
goose. For a historian this cries out for explanation: why are the forms
these and not foots and tooths and gooses? The answer, which was again
well known in Saussures day, is that, in an earlier period which is not
recorded, these nouns must have had a plural in -i. Such forms are
attested in the earliest records of German, with which English shares
a common ancestor: Old High German gast guest, plural gasti, or hant

14

Languages

hand, plural hanti. So, alongside the ancestral singulars fDt, tD^ and gDs,
which are attested in Old English, we reconstruct, for a prehistoric
period, the unattested plurals *fDt + i, *tD^ + i, *gDs + i. The forms we
actually have can then be explained by intervening changes. First the D,
in which the body of the tongue is raised towards the back of the mouth,
changes to B, in which it is raised towards the front. That is plausibly due
to the influence of the -i, in whose articulation the tongue is raised still
more towards the front. So, hypothetical *fDt + i, *tD^ + i, *gDs + i
become, still hypothetically, *fBt + i, *tB^ + i, *gBs + i. Then, still hypothetically, the -i is weakened to the point at which, in the end, it disappears; this gives us the forms attested in Old English: fBt, tB^, gBs. Finally,
from the later middle ages, the B or (in modern spelling) ee changed to, in
phonetic notation, [if].
This was and is, to repeat, a stock example; and it illustrates perfectly
how, from a historical linguists viewpoint, something that in the modern
language is a pure irregularity can be shown to make sense on the basis
of a reconstructed pattern. That is the insight, as we have seen, that
Saussure himself learned when he was young. But at any period in the
history of English what exists is a specific system, and such systems
(langues) are formed by the collective knowledge of their speakers. What
then is known to speakers of the present day? They know, of course, the
forms feet, teeth and geese; they know their meanings in relation to those
of foot, tooth and goose. Likewise, for example, speakers of Old English
knew the forms fDt, tD^ and gDs, and their relation in meaning to fBt, tB^
and gBs. In either period, both the forms as they are then and the relations in which they then stand constitute the language (la langue) that
exists at that time. The same is true of any period intervening between
Old English and the present; also of any earlier period in which, hypothetically, a set of speakers knew forms such as fDt and fDti or fDt and
fBti. But in no such stage do speakers know the forms that have been
current in earlier periods, or the relations of meaning that they entered
into, or the changes that have led from those forms to their own. Therefore these are not part of their langue, and a linguists description of the
langue of any body of speakers, at any stage in this history, deals in
irrelevancies if it says anything about them. For a langue is formed,
once more, by the collective knowledge of its speakers and these are
things that speakers do not know.
Linguists in general are therefore dealing with relations of two kinds.
There are relations, on the one hand, between forms belonging to la
langue as it is known in different periods: between earlier fDti and later
fBti, fBt or feet; or, for that matter, between fDt as reconstructed and Old
English fDt, or feet as known to speakers in the sixteenth century and

Linguistics as the study of language systems

15

present-day feet. These belong to the time dimension or, in Saussures

terms, to the axis of succession. His treatment of them was again within
the tradition in which he stood.
On the other hand, there are relations between forms that belong to
the same period: between present-day foot and present-day feet, between
Old English fDt and Old English fBt, and so on. These belong to the axis
not of time but, in Saussures terms, of simultaneity. Now for historical
linguists they are as they are because of changes that have given rise to
them. Synchrony, in Saussures terminology, is in that sense the product
of diachrony. But in a synchronic view English exists only in the collective knowledge of particular bodies of speakers. To say, for instance,
that fBt changes to feet, or that, by a sound law, B in general changes to
[if], is to presuppose langues that exist for different bodies of speakers in
different periods. To study diachrony is to establish relations between
successive synchronies.
A diagram in which these axes are contrasted (115) has often been
reproduced. But their absolute divergence is best illustrated by a comparison, which in the Cours itself is claimed to be more revealing than
any other, between the history of a language and a chess game. Between
moves chessmen form a specific pattern on the board, in which the relation of each piece to the whole depends on its position and the positions
of the others. This corresponds to a specific state of the language (un
tat de la langue, 125): that is, to a state of synchrony. But each such
state is strictly temporary. In the game a player moves just one piece
at a time, but that one change results in a new equilibrium, or new
synchrony. It has a repercussion on the entire system (un retentissement sur tout le systme), and it is impossible for the player to foresee
all its effects (126). Likewise, in the language, each change affects only
isolated elements; but again a new synchrony, as a whole, replaces the old
synchrony. The movement of a piece is, finally, a fact entirely distinct
from the states of equilibrium that precede and follow it (un fait
absolument distinct de lquilibre prcdent et de lquilibre subsquent).
Nor, in considering a particular state of the board, does an observer need
in general to know the sequence of moves that have led to it. All this,
we are told, applies equally to the language, and confirms the radical
distinction of what is diachronic and what is synchronic (Tout ceci
sapplique galement la langue et consacre la distinction radicale du
diachronique et du synchronique, 128). The only difference is that changes
in a language are not intentional, while moves in the game are. In other
respects, a game of chess is like an artificial realisation of what the
language presents in a natural form (est comme une ralisation artificielle
de ce que la langue nous prsente sous une forme naturelle, 125).

16

Languages

First, then, the study of language in general (langage) is divided into

those of languages (langues) and of speech (parole). The former is
linguistics in the strict sense. Secondly, the linguistics of languages is
in turn divided into diachronic linguistics and synchronic linguistics.
These are equally valid but diverge absolutely. Our third point then concerns the notion of a language as a system of values. In economics, for
example, there are systems of values that relate work to the wages that
are paid for doing it. Likewise (for Saussure) the values of words are their
meanings. In each case we are concerned with a system of equivalences
between things of different orders (il sagit dun systme dquivalences
entre des choses dordres diffrents, 115). In one it is between work and
wages. In the other it is between a signifiant, or something that means,
and a signifi, or something that is meant.
To understand the relation between these it may help if we begin with
another analogy. In the system of coinage that is in use in Britain as I
write, a piece of metal of a certain alloy, size and design has, for example,
the value of ten pence. It is one of a determinate set of coins with different values, and these values are interrelated. For example, ten coins with
the value of ten pence are equivalent to one with the value of one pound.
The system holds only for a specific society at a specific time. There is no
independent reason why this particular piece of metal should have that
particular value, or why this particular set of values should be involved.
There was a period in which a larger coin, which had among other things
the words two shillings on it, also had the value of ten pence. There was
a period in which a coin with sixpence on it had the value of two and a
half pence; now no coin has that value. Earlier still there was a period in
which the values were in shillings and pence, with twelve pence to the
shilling and twenty shillings to the pound, whereas now they are decimal,
with a hundred pence to the pound. Coins are coins, moreover, solely in
that they have values. A piece of metal which in many respects resembles
a coin is no more than a piece of metal if it has no value in the system.
Thus an object that was once a sixpence is, in the present system, a coin
no longer. Equally, two bits of metal that are partly different may, in a
specific system, be the same coin. Thus a one-pound coin may have
a thistle on the back and, round the side, the Latin inscription nemo
me impune lacessit. Alternatively, it may have the royal arms and the
inscription decus et tutamen. Alternatively, it may have a leek and the
Welsh inscription pleidiol wyf im gwlad. But each of these is equally a
one-pound coin. In all these ways the system is a matter of convention,
again in a specific society at a specific time.
The example of coins is used only briefly in the Cours (160), and not
for this purpose. But it is easy to see how the analogy would apply. A

Linguistics as the study of language systems

17

language is a system of signs relating signifiers and things signified;

and, just as a coin has its value only in a specific system of coinage, a
signifier has its value only in a specific language system. Outside the
system of coinage a coin is a mere piece of metal. Outside the language
system a word would likewise be nothing. It is a word precisely because it
has a value; and, just as the value of coins is in relation to those of other
coins, so the value of words is in relation to those of other words.
This is a bold conception, and its full originality may not at once be
clear. The notion that words are signs was not, of course, new. Derivatives of the Greek word for sign (sBmeon) were part of the terminology
of the Stoic philosophers in the third century BC; terms like signify and
signification are from the Latin word corresponding to it; and among
Saussures more immediate predecessors, the American linguist William
Dwight Whitney writes straightforwardly of a language as an aggregate
of articulated signs for thought (1867: 22), or a system of signs (1971
[1875]: 115). Nor was it new to insist that signs in language were matters
of convention. In Saussures terms they are arbitrary; and, as Eugenio
Coseriu has shown, this term derives ultimately, through various changes
of formula, from a concept of convention in Plato and Aristotle (Coseriu,
1967). The Cours contains few acknowledgments of Saussures predecessors. Again we must remind ourselves that it was in origin a course of
lectures, in days long before handouts. But for this point Saussure specifically referred to Whitney (Cours, 110). In the second of the passages I
have cited Whitney made clear that the relation between the sign and what
it signifies is one of mental association only. Each individual form of
human speech is a body of arbitrary and conventional signs for thought
(1867: 32). For any individual who learns to use it, every vocable is first
arbitrary, because any one of a thousand other vocables could have been
just as easily learned . . . and associated with the same idea; secondly,
it is conventional, because it ha[s] its sole ground and sanction in the
consenting usage of the community of which it formed a part (14).
In part, then, what the Cours says was familiar and ancient. But other
things were not, and the notion of values is central to them. The most
obvious, which I have consciously glossed over, is the way in which
signs are defined. For Whitney, for example, the sign was the word or
vocable itself. Each word is a sign for thought and, as such, stands in a
relation to an idea or concept. For the Cours, the signifier is an acoustic
image in the minds of speakers; what it signifies is a concept that, as
we saw earlier in this section, is triggered by it. But the linguistic sign as
such encompasses both. It is a mental entity with two sides (une entit
psychique deux faces): one a signifiant or signifier, and the other a
signifi or what is signified (99).

18

Languages

This may at first sight seem perverse. When we say, for instance, that a
black cloud is a sign of rain what we are calling a sign is precisely the
black cloud. In the same way, if we are told that the word tree, or an
acoustic image corresponding to tree, is a sign in our minds of the concept tree, we might normally take the sign to be the word or acoustic
image itself. It is perhaps not surprising, therefore, that many writers
who talk of signifiants and signifis in fields other than that of language,
while appealing ostensibly to Saussures theory of larbitraire du signe,
in reality revert to ordinary usage. But the reason for defining the sign in
the way he did is that neither side of it exists independently. The cloud is
one thing, the rain another: we can identify each as a separate entity. But
in the case of language we cannot point to a signifiant independently of
its signifi, or to a signifi independently of its signifiant. A linguistic
entity exists only through the association of one with the other (Lentit
linguistique nexiste que par lassociation du signifiant et du signifi,
144). Thus, on the one hand, an acoustic image tree is identified solely
by its relation to the concept tree; and, on the other hand, a concept
tree is identified solely by its relation to the acoustic image tree. The
sign as a whole, moreover, is identifiable solely by its relation to other
signs, within the system of values of which it is one term.
Let us begin with the signifi. An ancient view, which was still current
enough in Saussures day, is that words are names for things. Trees are
things of one kind, stones things of another kind, and in each language
there are names, like tree and stone, for each of them. The Cours dismisses this view from the outset: a language is not a nomenclature (97).
Various reasons are given, but the point that most concerns us is that
there is no preexisting set of meanings. From a psychological viewpoint,
taken in abstraction from its expression by words, our thought is simply
a shapeless and undifferentiated whole (Psychologiquement, abstraction
faite de son expression par les mots, notre pense nest quune masse
amorphe et indistincte, 155). Until the language system comes into play
(avant lapparition de la langue), nothing is distinct.
It is no better with the signifiant. Traditionally, what signifies is part
of a physical signal, and the Cours speaks similarly of a slice of sound
(une tranche de sonorit, 146). But how do we identify recurring slices?
They are not physically marked off: in a signal part of which we can
represent in writing as cut trees, the t of cut and the t of trees will run into
one another, and the s of trees, which in a plural like this has a meaning
of its own, would be no more a distinct slice than the ze of seize, which
does not. A slice, in short, is a slice only by virtue of its relation to what
it signifies. Nor are recurrences of the same slice physically constant.
The Cours gives the example of a lecture in which the word Messieurs!

Linguistics as the study of language systems

19

Gentlemen! is repeated several times. As we perceive it, it is always the

same expression: but the speakers delivery and intonation will vary so
much that the physical differences between one instance and another will
be as great as those between words that we perceive as different, such as
paume palm of hand and pomme apple (150f.). It is not, then, as a
physical entity that Messieurs remains the same word.
To make this point, the Cours gives among others a famous analogy
between the sameness of a word and the sameness of a train in a
railway system. The 8.45 p.m. express from Geneva to Paris leaves one
day, and the same train leaves again twenty-four hours later. But, physically, it too is not the same: it probably has a different locomotive, different carriages, a different driver, and so on. What makes it that particular
train is its time of departure, its route and in general all the circumstances that distinguish it from all other express trains (et en gnral
toutes les circonstances qui le distinguent des autres express, 151). The
identity of words is said to be like that. What makes each instance
of Messieurs! the same word Messieurs is its place within a system in
which, just as one train is distinguished from another within the railway timetable, the sign that it represents, which is again formed by the
association of a signifiant with a signifi, is distinguished from other
signs.
We have now reached the conclusion to which Saussures thought had
been tending. In the beginning, I equated meaning and value; and,
when the Cours first talks of linguistics as a science that deals with values,
it too does so (115). But, as a later section makes clear (158ff.), value is
a matter of difference. The value of a ten pence coin lies in its relation to
the values of others within a decimal system. What makes the train the
8.45 express is the distinction or opposition between it and other trains in
the timetable: its time of departure as distinct from that of other trains
from Geneva to Paris, its status as an express in opposition to that of a
stopping train, and so on. Likewise a linguistic sign exists by virtue of its
opposition to other signs; and, just as coins have values only within a
particular system of coinage, and the identity of trains is only in terms of
a particular railway system, so the links established between signifiants
and signifis exist only through the system of oppositions by which,
literally, that particular language is formed. The conclusion is stark and
radical. For what we have said amounts to the statement that in the
language system there are only differences (Tout ce qui prcde revient
dire que dans la langue il ny a que des diffrences). No unit has an
existence independent of the relations between it and other units. In the
language system there are only differences, with no positive terms (dans
la langue il ny a que des diffrences sans termes positifs, 166).

20

Languages

We now have one structuralist answer to the question with which we

began this chapter. A language, such as English, is a social phenomenon (fait social) whose existence is solely as a system of values holding
in a specific society. Its manifestations through speech (parole) are individual and fleeting. Therefore, in the first instance, it is the system (langue)
that we must study. When languages change, one system is replaced by
another. Therefore the study of changes must be distinguished rigorously
from the study of a system in abstraction from time. Each language, as
the Cours insists at this point, is a system of pure values (un systme de
pures valeurs), determined by nothing other than the momentary state
of its terms (que rien ne dtermine en dehors de ltat momentan de ses
termes, 117).
How far was this account accepted? The second of our three points has
been widely taken for granted. By the middle of the twentieth century
many linguists were to deal with synchrony exclusively. But not everyone
has followed Saussures theory of la langue. Languages may be the
primary or a primary object of study. But their relation to speech is not
as the Cours said it is.
2.2

Languages as sets of utterances

Let us begin with one explicit critic. The British linguist J. R. Firth was a
structuralist in many ways. From the 1940s, when he founded the first
department of linguistics in the University of London, he was certainly
a leading member of the group that those who were not structuralists
opposed. He also agreed with Saussures Cours on some points. In a
famous passage, Saussure had envisaged a new science of semiology
(smiologie) that would study the life of signs in the context of social
life (la vie des signes au sein de la vie sociale) (Cours, 33). Its name is
again from Greek sBmeon sign; it would be part of social psychology
and would have linguistics, in turn, as one subdiscipline. To Firth, writing in the 1930s, this was perhaps the most striking thing in the whole
of de Saussures great work: that linguistics can only find a place among
the sciences if it is brought into relation with semiology (Firth, 1957
[1935]: 17).
But for Firth the study of language was that of concrete instances
of speech. Such instances are social events, and the reality of language
lies in the social behaviour of speakers. Hence he rejected Saussures
concepts of langage, langue and parole. For Saussure, in Firths words,
languages were social facts conceived as sui generis and external to and
on a different plane from individual phenomena (Firth, 1957 [1950]:
179). [T]rue Saussureans, like true Durkheimians, regard the structures

Linguistics as sets
Languages
the study
of utterances
of language systems

21

formulated by linguistics or sociology as in rebus (181). The structure is

existent and is treated as a thing. For Firth this made the individual
speaker, in his terms, an underdog . . . whose speech was not the integral and concrete object of linguistics (183, translating Cours, 23). He
thus rejected what he calls, in the same sentence, de Saussures mechanistic structuralism. The systematics of specific aspects of linguistics, of
the study of sounds, of grammar, and so on, do not describe an underlying reality, stored in the collective conscience. They are ordered
schematic constructs, frames of reference, a sort of scaffolding for the
handling of events (181). Firth does not refer at this point to Saussures
grand vision of semiology. But the reality of language is precisely in la
vie sociale, the social life, of a community of speakers.
This does not tell us what we might mean by a language, and perhaps,
if we take Firths view, that issue is not central. But one answer, not his
but one that has been very influential, is that it is the sum of the speech
that is possible in a community. Anything said by a speaker is, in linguists terminology, an utterance. A language, therefore, is a body of
utterances.
This was the answer given, in 1926, in a remarkable article by Leonard
Bloomfield. He was then in his late thirties, and had already published a
general introduction to linguistics (1914) influenced especially by ideas in
Germany at the turn of the century. But by the early 1920s he had read
Saussures Cours, whose second edition he reviewed in 1924. At the same
time he was converted to what he later calls a physicalist philosophy, in
which all statements in science are seen as ultimately about physical events.
Its immediate inspiration lay in the new school of behaviourist psychology: in particular, in the ideas of Albert P. Weiss, who was a colleague
in the 1920s at Ohio State University. But such a philosophy was also
developed independently, in this period, by the logical positivists in
Vienna. The Vienna school had found, as Bloomfield put it later, that
all scientifically meaningful statements are translatable into physical terms
that is, into statements about movements which can be observed and
described in coordinates of space and time (Bloomfield, 1970 [1936]:
325). Statements which cannot be made in these terms are either scientifically meaningless or else make sense only if they are translated into
statements about language. By the 1930s, one could not read modern
writings without meeting such ideas again and again (323); and, in the
future testing of this hypothesis of physicalism, linguists will have to
perform an important part of the work.
It is in this philosophical setting that we have to read both Bloomfields
Language (1933) and his important article in 1926, which is in part a
precursor. Its title is modelled on an essay by Weiss, which had sought to

22

Languages

make explicit basic postulates that underlay behaviourism in psychology (Weiss, 1925). In a far more detailed study Bloomfield tried to formulate, as rigorously as possible, a set of postulates for the science of
language. A postulate is an axiom or assumption that we have to take
as given. For example, by Bloomfields Assumption H1, with which he
begins a section on historical linguistics, we take it as an axiom that
every language changes at a rate which leaves contemporary persons
free to communicate without disturbance. Such postulates rest, in part,
on terms which are defined within our science: thus Assumption H1 rests,
crucially, on the prior definition of a language. Then, given the postulate, further terms can be defined. Thus the case in which communication
is disturbed but not destroyed by change defines a dialect (Bloomfield,
1970 [1926]: 136). The advantage of the postulational method is that
it forces us to state explicitly what factual assumptions we are making,
to define our terms precisely and show what depends on what (126).
What do we have to assume in order to talk of languages? Clearly, a
language is that of a community, whose members communicate with one
another by it. Our starting point might be that there are such systems of
communication, langues as Saussure conceived them. But Bloomfields
strategy is more subtle, and begins with the insight that, in such communities, there must be similarities between different acts of speech. They
are similar because, in ordinary terms, the members of the community
are speaking the same language. So, it is through such similarities that
languages are established.
By Bloomfields first definition, an act of speech is an utterance. Our
crucial assumption, therefore, is that within certain communities successive utterances are alike or partly alike. Suppose, to follow Bloomfields
illustrations, that a stranger comes to the door and says Im hungry.
Then suppose that, on another occasion, a child who has eaten but does
not want to go to bed says Im hungry. We will take it, as linguists, that
these two acts are alike: that the sounds the speakers utter are the same,
with the same meaning. There can also be likenesses at individual points.
If someone said The book is interesting that would be one utterance;
Put the book away would be another. But for a linguist they are partly
alike, since both include the book. Any community in which there are
such resemblances is by definition a speech-community; and, in such a
community, a certain range of utterances will be possible. Thus, in the
community in which one can say The book is interesting, one can also
say Im hungry but not, for example, what would in another speechcommunity be the utterance (in French) Jai faim. The language of [a]
speech-community is accordingly the totality of utterances that can be
made in it (Bloomfield, 1970 [1926]: 12930).

Linguistics as sets
Languages
the study
of utterances
of language systems

23

This rests, to repeat, on an assumption, which is axiomatic for a particular science. What the stranger at the door says is objectively very
different from what the child says. Their voices are pitched differently,
one may speak softly and the other loudly, and so on. The actions that
accompany their speech will also differ. The stranger stands still on the
doorstep; the child might be vigorously resisting a parent who is trying to
carry it upstairs. One may indeed be hungry, and their purpose is to ask
for food. The other is not hungry at all. Outside our science, as Bloomfield
says, the similarities are only relative. Only within it are they absolute
(again 129f.).
But without this fiction, as Bloomfield also calls it, linguistics would
be without foundation. We say, for example, that both speakers utter a
word hungry, and that this word has, in each case, the same meaning.
That the child is lying and the stranger telling the truth is not, for us as
linguists, relevant. But how can we justify this? The Saussurean answer is
that hungry is a sign uniting an acoustic image hungry with a concept
(hungry); but, unless we can identify these physically, that answer is
now closed to us. We simply have to assume that such relations exist. In
Bloomfields Language this is the fundamental assumption of linguistics.
In principle, our statements could be founded on the discoveries of other
sciences, regarding physical states of the brain or other parts of a speakers
body. But, failing that, we have to assume that, in certain communities
(speech-communities) some speech-utterances are alike as to form and meaning (Bloomfield, 1935 [1933]: 144). The point is italicised twice (see also
78). Only on this basis can we say that, in the specific case of hungry, the
form that is phonetically [hkpgri] has, in both speech-utterances, the
same meaning hungry.
Bloomfield did not, in 1933, repeat his earlier definition of a language.
But let us take English, for example, to be the totality of utterances
possible in English. To describe English is thus to describe these utterances, and the structure of English will accordingly be the structure
that, taken as a whole, they have. What exactly, then, is an utterance?
For Bloomfield, it was an act of speech. This is again from 1926; in
Language utterance is not treated or indexed as a technical term. But, as
Bloomfield makes plain, the act is that of emitting the sounds that
constitute speech. A speech event is described in strictly physical terms:
the movements of a speakers vocal cords and other organs; the resulting
sound-waves; their action on a hearers ear-drums (Bloomfield, 1935 [1933]:
25). An act of speech is therefore distinguished carefully from whatever
surrounds it, in the state of the world or in a speaker or hearers other
behaviour (23). An instance of speech is also referred to, in passing (23), as
a speech-utterance. The utterances that we describe are thus specifically

24

Languages

the sounds that speakers emit. The totality of utterances that can be
made in a community is the totality of sounds that, in speaking their
language, members of that community may emit. To describe the language is, accordingly, to describe the structure of such sounds.
This leads directly to the programme developed, largely after
Bloomfields incapacity and death in the 1940s, by his follower Zellig
S. Harris. In one of his earliest writings as a structural linguist, Harris
rejected the notion that a language structure, conceived as we will see
in the Saussurean tradition, can be studied independently of the speech
act. The former, he says, is merely the scientific arrangement of the
latter (Z. S. Harris, 1941: 345). With this remark Firth, who we cited at
the beginning of this section, might well have agreed. But for Harris what
we were studying was not, in the first instance, the entire event of which
speech is a part; it was again the sounds uttered. In describing English we
establish that, for example, the sounds represented by Im hungry constitute a possible utterance. We analyse them into recurring units: for
example, a unit I recurs in other utterances such as I must eat or You know
I wont. We establish relations among units: for example, in each of these
utterances, I bears the same relation (traditionally as subject) to what
follows; in Im hungry and Shes clever, [a]m and hungry are related to
each other in the same construction (as it is traditionally called) as [i]s
and clever. We group units into classes: thus, by virtue of such relations,
hungry and clever are both (again as they are traditionally called) adjectives. In this way we assign a structure to each utterance, by which it is
again classed with others. Thus Im hungry has a structure partly shared
with I must eat, in which a subject is related (still in traditional terminology) to a predicate ( [a]m hungry, must eat). The predicate has a structure
in which a certain class of verb ( [a]m) is related to an adjective (hungry);
that is also the structure of [i]s clever, and so on.
The method is explored in Harriss first book as a general linguist
(1951). A language or dialect is, in general, the talk which takes place in
a language community (13). Thus, in a particular case, it is the speech of
the community (9) selected for study. Our data are instances of such
talk, and any particular body or corpus of data is a sample of the talk
that is possible. The description of the language structure, as Harris
calls it in his final chapter (372f.), is an account precisely of the structure
that, when we apply this method, talk is found to have.
One further insight remains, whose most important consequence we
will reserve for a later chapter (6.1). Briefly, however, a language is in this
view a set. Its members are utterances; and, at the level of abstraction at
which we are now talking, each utterance is characterised by the relations
established over a set of smaller units. Although we are talking, in principle,

Linguistics
The
autonomy
as the
of study
linguistics
of language systems

25

of sounds that speakers may emit, we have in effect abstracted something

very like a set of written sentences, in which determinate words stand in
determinate orders. The set as a whole is also, by implication, determinate.
We have talked of utterances that can be made in a community; so, at
this level of abstraction, certain combinations of smaller units are members of our set. Others, we imply, are not members. What then are the
properties of the sets that we call languages, that distinguish them, in
general, from other sets? In abstraction from reality, sets are mathematical objects. Our question, therefore, is about the properties of languages
as mathematical systems.
This insight did not occur to Bloomfield in the 1930s, and he might
well have recoiled from it. But it was to inspire Harris for the remainder
of his life, which ended in his eighties, a year after his last book (Z. S.
Harris, 1991). Meanwhile it had led directly, in the 1950s, to the pivotal
concept of a generative grammar. A set is generated if, by following a
series of rules or instructions, its membership is specified. For example,
the instruction multiply x by 2, where x is a whole number generates the
set of even numbers. A language is, in this view, a set; its members
are utterances or, equivalently for Harris, sentences. A grammar is
traditionally a description of a language, and could therefore be viewed,
as Harris puts it in one paper, as a set of instructions which generates the
sentences of the language (1954: 260).
This idea was in the event developed not by Harris himself, but by his
pupil Chomsky. It is indeed a pivot on which what is often called the
Chomskyan revolution turned. But we will need to remind ourselves again
of its Bloomfieldian ancestry.
2.3

The autonomy of linguistics

For the moment, we must return to Bloomfield. We have seen that his
technical concept of a language differed from Saussures. But on what
were then essentials they more often agree. In a letter written in the 1940s
Bloomfield said that his Language reflected Saussures Cours on every
page (see Cowan, 1987: 29). To understand the significance of either, we
must try to grasp what he meant. In the early 1920s Bloomfield wrote
already of a newer trend of linguistic study, to which the Cours gave
a theoretic foundation (Bloomfield, 1970 [1922]: 92). It affects, he said,
two critical points. The first concerns the status of historical linguistics.
We are coming to believe, as Bloomfield put it, that restriction to historical work is unreasonable and, in the long run, methodically impossible.
Synchronic study, in Saussurean terms, is prior to diachronic. I am citing
from his review of another general introduction to language, published in

26

Languages

1921, by his American contemporary Edward Sapir (Sapir, n.d.).

Bloomfield is glad to see that Sapir deals with synchronic matters . . .
before he deals with diachronic. The same concern was to dictate the
structure of his own book, which has a clear division between chapters
on the nature of language, roughly its first half, and those on the history
of languages. There was at least one earlier precedent, in the work by
Von der Gabelentz cited at the beginning of this chapter (Gabelentz,
1901 [1891] ). But Saussures analysis was from now on seen as seminal.
The second critical point concerns the relation of linguistics to psychology. Bloomfields first book (1914) had been founded in the psychology
of Wilhelm Wundt, who had himself published two thick volumes on
language at the turn of the century (Wundt, 191112 [1900] ). Two decades
earlier, Hermann Pauls great work on the principles of language history
had relied in part on an earlier psychological system (Paul, 1920 [1880] ).
At the time it was hard to write on the theory of language without reference to either. A monograph by the Indo-Europeanist Berthold Delbrck
is in substance an extended review of Wundts work, in relation, in particular, to Pauls (Delbrck, 1901). But its first chapter is devoted to the
underlying psychological systems, on the grounds that, only if they were
first understood, could either view of language be assessed (iii).
Delbrck acknowledged that, for practical purposes, it does not matter
what view of psychology a linguist holds (fr den Praktiker lt sich
mit beiden Theorieen leben, 44). But in Bloomfields newer trend it did
not matter for any purpose. [W]e are casting off , in his words, our
dependence on psychology, realizing that linguistics, like every science,
must study its subject-matter in and for itself, working on fundamental
assumptions of its own (again Bloomfield, 1970 [1922]: 92). Bloomfields
Language was conceived as a revision of his first book. But in the interim,
as he explains, we have learned . . . what one of our masters suspected
thirty years ago, namely, that we can pursue the study of languages
without reference to any one psychological doctrine. To do so safeguards our results and makes them more significant to, among others,
psychologists themselves (Bloomfield, 1935 [1933]: vii). In psychology,
Bloomfield was by then a behaviourist, in keeping with his view of science generally. But, as a linguist, he was practising another discipline. It
rested on an assumption, as we have seen; but once this was made it
would be autonomous.
It will help at this point if we look, in particular, at the problem of
defining sentences. Suppose a speaker says Im hungry. Cant we have
lunch? For grammarians two sentences are uttered: first Im hungry;
then Cant we have lunch? But what distinguishes them, and what is the
nature of this unit?

Linguistics
The
autonomy
as the
of study
linguistics
of language systems

27

In the traditional definitions, speakers express thoughts. They do so

by combining words; so, in one of the formulae inherited from late antiquity, a sentence or utterance (Latin oratio) is a coherent ordering of
words (ordinatio dictionum congrua) expressing a complete judgment
(sententiam perfectam demonstrans) (Priscian in Keil, 18559, I: 53).
Definitions like this had been accepted as they stood for centuries. But
thoughts or judgments are formed in the mind, and psychology is the
science of the mind; therefore psychologists were bound to speculate
about the process. Does the thought arise first as an unanalysed whole?
If so, it must be subject to a process of analysis; in this way it is divided
into logically articulated parts; the sentence as spoken is an expression of
that. That was, simply and perhaps crudely, the account proposed by
Wundt (II: 242ff.). Alternatively, the thought arises not as a whole, but
from the linking of concepts that are initially separate. That was the
basis, though I have once more simplified it, of the definition by Paul (121).
Throughout this period theories of the sentence multiplied. Although
the traditional account raised other problems, it was hard to see how the
formation of thoughts in the minds of speakers, and their corresponding
perception in the minds of hearers, were not central.
The treatment in Saussures Cours is notoriously inexplicit. But if
sentences express thoughts, and thoughts arise in individual minds on
individual occasions, the sentence cannot be a unit of the social fact that
constitutes a language. That indeed is what, at one point, the Cours
says: [la phrase] appartient la parole, non la langue (172). Otherwise
it says almost nothing; unusually, for a book of its date. But, if this interpretation is right, the linguistique de la langue had no concern with it.
What did concern it were the relations in which, within Bloomfields
utterances, successive words stood. Suppose again that a speaker says
Im hungry. This act of speech arises, in part, from a free choice of
words: the speaker could alternatively have said Im starving, or, in
different circumstances, Were sorry or How disgusting! In Saussures
terms, such acts of speech again belong to parole. But the way in which
words are combined is, in part, regular: [a]m regularly agrees with I as
[a]re agrees with we, both regularly precede words such as hungry or
sorry, and so on. Such relations must belong to the language system.
They are part of what is shared by members of a community speaking
English, just as, to return to the analogy of 2.1, the way that knives are
held in one hand and forks in the other, again on individual occasions
when people are eating, belongs to a shared system of table manners.
Relations like these are of a kind that the Cours describes as
syntagmatic (172f.). Unfortunately it says very little about them, and
any interpretation beyond this point is pure extrapolation. But if such

28

Languages

regularities exist, there must, in any utterance, be combinations of words

that conform to them. If someone simply says Im hungry, this combination is by implication the whole sequence I + am + hungry. But
suppose again that they say, first, Im hungry and then, without pause,
Cant we have lunch? This too is a single act of speech; in Saussurean
terms a single instance of parole. Now there are further regularities that
constrain the combination can + not + we + have + lunch, which is our
grammarians second sentence. But nothing in the language system
will constrain the link between it and the first. A speaker could equally
well say Cant we have lunch? Im hungry, or, in the same or other
circumstances, Im hungry. Stop the car, Im hungry. Why cant we
find a restaurant?, and so on. Within sentences relations are part of the
language system. Across sentences they are not.
This was effectively Meillets definition, in a manual of Indo-European
that preceded the Cours by four years. It is formulated purely from the
linguistic viewpoint (au point de vue linguistique) in abstraction from
anything to do with logic or psychology (et abstraction faite de toute
considration de logique ou de psychologie). The sentence is therefore
no more than a complex of articulations bound together by grammatical
relations (un ensemble darticulations lies entre elles par des rapports
gramaticaux) which are self-sufficient in that they do not depend grammatically on any other complex (et qui, ne dpendant grammaticalement
daucun autre ensemble, se suffisent elles-mmes) (Meillet, 1937 [1912]:
355). If we think back, that was all that Delbrcks practising linguist
needed. But by the 1920s it had a clear theoretical foundation, in a linguistics that does not rely on either of the psychological theories with
which, at the turn of the century, Delbrck had been obliged to grapple.
The final, classic formulation was Bloomfields. In his first book he had
followed Wundts definition (Bloomfield, 1914: 60). But in the 1920s he
accepted Meillets, contrasting its simplicity and usefulness (Bloomfield,
1970 [1926]: 129, n. 6). All that remained was to ground it directly in his
fundamental assumption. As we have seen, Im hungry is partly alike
in form and meaning, under that assumption, to, for example, I must
eat or You know I wont. But utterances are not merely alike to
others with which they share words. Under the same assumption, Im
hungry is also alike to other utterances with which it shares its construction. Thus, though the words are different, their arrangement and the
meaning associated with it are like those in, for example, Mary was
furious or Were sorry. Im hungry is accordingly, by definition, a
linguistic form: a form that, as a whole, has a meaning (Bloomfield,
1935 [1933]: 138). So too, by a similar analysis, is Cant we have lunch?
Sentences can then be defined quite simply as linguistic forms that are not

Linguistics
The
autonomy
as the
of study
linguistics
of language systems

29

included in any larger linguistic form (170). In our speakers utterance,

no specific meaning is associated with the way in which these are in turn
arranged. Therefore there is no larger linguistic form Im hungry. Cant
we have lunch? Therefore this utterance includes two sentences, not one.
The trouble over the nature of the sentence had been largely
nonlinguistic (again Bloomfield, 1970 [1926]: 129, n. 6). If linguists
simply study their own subject matter, in and for itself , such problems
are solved. The in and for itself is Bloomfields (again 1970 [1922]: 92).
But it could just as well be from the famous final sentence of Saussures
Cours. It is one that, as it happens, was not in his students notes.
But even if it is the editors own, it makes clear what Saussure could well
have called his fundamental idea, that the only true object of study in
linguistics is the language system, considered in and for itself (lide
fondamentale de ce cours: la linguistique a pour unique et vritable objet
la langue envisage en elle-mme et pour elle-mme) (Saussure, 1972
[1916]: 317).
It is in this sense that Saussures influence is on every page of
Bloomfields own book. But the autonomy of linguistics had clearly been
achieved at some cost. In particular, we have to assume, as Bloomfield
makes clear, that relations between forms and meanings, both in vocabulary and grammar, are constant. Only on that basis could we speak of
langues or languages, and only then could a linguistics of the language
system, to translate once more the formula of the Cours, be founded.
The problem Bloomfield and Saussure both faced is that, when people
talk, no utterance is exactly like any other. Someone says Im hungry;
the next day, as we will describe it, they say again Im hungry. But each
event is transient, and the sounds emitted will not be the same, in minute
or even gross respects. A woman with a bad cold asks How should I
know? A tenor sings (from Brittens Curlew River) How should I know?
The sounds emitted are in that case very different. Nevertheless, we will
describe these as the same words. On what basis can we do so?
The answer for Saussure lay in the system of values that constitutes a
language system. In it, for example, an acoustic image, such as know, is
linked indissolubly, as the signifiant that forms one aspect of a linguistic
sign, to a concept (know), as the signifi corresponding to it. But
Bloomfields answer is in essentials the same. For acoustic image read
form, which is defined, in his set of postulates, as a recurring vocal
feature (Bloomfield, 1970 [1926]: 130). For concept read meaning.
But, by his fundamental assumption, forms and meanings are again
indissolubly linked. For it is only through the recurrence in tandem of
both vocal features and features of meaning, of a distinctive vocal feature
that is present whenever someone says know and a distinctive meaning

30

Languages

feature that is present whenever someone does so, that a permanent

linguistic unit exists.
For Bloomfield meanings were, quite literally, present. As objects of
scientific study they had, in principle, to have a physical identity. The
meaning of an utterance was accordingly defined by the external circumstances, each describable in space and time, in which the act of speech
takes place (Bloomfield, 1935 [1933]: 27). This view was at the time
so radical, and differed so clearly from that of anyone who took thoughts
or concepts in the mind for granted, that it is on Bloomfields general
philosophy of meaning, often linked specifically to behaviourist psychology, that most commentators dwell. But the point that now concerns
us is the link between a meaning, however described, and the form
associated with it. Under Bloomfields fundamental assumption, each
term in this relation again presupposes the other. For the recurrence of
one is only by virtue of the parallel recurrence of the other; and, in
Bloomfields treatment, which is thought through more than that of
Saussures Cours, this is true of the arrangements in which recurring
parts of utterances stand, and the constant meanings which they, in their
turn, must be assumed to have, not just of the parts themselves. We will
see later how this view could be developed.

The prehistory of the phoneme

31

Sound systems

The chief, if not the only, interest of the Cours de linguistique gnrale
was as an account of the foundations of linguistics. Of its five numbered
parts, the last three deal conventionally with sound change and changes
in grammar, the distribution in space of languages and dialects, the reconstruction of prehistoric languages, and other topics natural in a manual
of its day. These together form more than a third of the whole (193317).
The chapters everyone now cites are part of the introduction, most
of Part 1 (General Principles) and most again of Part 2 (Synchronic
linguistics), a third again. But the earliest reviewers, as Keith Percival
has shown, did not see in these the revolution that was later proclaimed.
The book was seen more as old-fashioned (Percival, 1981).
That is perhaps not so surprising. For work on the foundations of
a discipline need not have immediate repercussions on the way it is
practised. When the Cours appeared, most linguists worked on IndoEuropean or some other family, on the history or grammar of particular
languages, in dialectology, and generally in fields to which it offered
nothing new. Even a synchronic linguist could learn little. The treatment of speech-sounds, for example, was based on lectures given in 1897
(editors note, 63) which were already dated. For the rest, we might be
tempted to recall a remark by Delbrck cited in an earlier section (2.3).
Provided that their methods are not disturbed, practising linguists can
live happily with whatever any theorist says about the philosophical
principles that underlie what they are doing. Only other theorists, of
whom there are at any time few, need respond.
For whatever reason it was not until the 1930s that Saussurean
structuralism took off. To understand both why and how it did we have
to look especially at the emergence of the basic unit called the phoneme.
The term itself (originally French phonme) had been coined innocuously in the early 1870s for a single speech-sound (German Sprachlaut).
At the time it was a new term and no more. But by the 1930s it was the
centre of a new and well-developed theory. Even historians of languages
could not ignore it: for sound change, as Bloomfield among others made
31

32

Sound systems

clear, was change in phonemes (Bloomfield, 1935 [1933]: 351). By the end
of the decade there were techniques by which, in any language, phonemes
could be identified. Ten years later, as we will see in a later chapter,
methods modelled on them were affecting the description of a language
generally. Finally, the theory fitted beautifully with what the Cours had
said about the object of the discipline.
3.1

The prehistory of the phoneme

The strands that led to it are complex, and we cannot survey every contribution. For those I will omit see especially the penetrating history of
this field by Stephen Anderson (Anderson, 1985). From the beginning,
however, we can distinguish two main ideas.
The first is that of distinctive differences among sounds. English blather,
for example, has a consonant distinct from that of bladder. The sound
written dd is produced by a temporary blocking of the outgoing airstream. In phonetic terminology, it is a stop or plosive (in notation [d] ).
The one written th is a fricative, produced with a turbulent airstream
but without blocking it (in notation, [2] ). The words are different; and,
in general, any word with [2] or [d] in this position ( feather, say, or body)
would be altered if the other were to replace it (fedder, bothy). The
classic theory dates again from the 1930s. But in its terms [2] and [d] are
different phonemes, characterised by distinctive features which in one
case might include a feature fricative and in the other an opposing feature plosive. In Saussurean terms, distinctiveness is one of the relations
that define a language system. So, in the system we call English, fricative
and plosive are two elements.
The second idea is that units of sound alternate. In English, for example, the diphthong [k0] in drive ( [drk0v] ) is said to alternate with [iv]
in drove ( [drivv] ) or [0] in driven ( [dr0vin] ). These are forms of the same
verb (to drive); but, while the dr . . . v is constant, the vowels intervening
vary. For historical linguists this is an irregularity which has to be explained by older stages of the language and the changes that have ensued;
and, for such alternations in English, their account is varied and complicated. But the pattern that has resulted is in part systematic. As drive is
to drove so strive is to strove and ride is to rode; and, for some speakers,
dive is to dove. As sing ( [s0p] ) is to sang ( [sap] ) and sung ( [skp] ) so ring,
for example, is to rang and rung or drink (with final -nk) to drank and
drunk. As present tense read ( [rifd] ) is to past tense read ( [rtd] ) so lead is
to led, meet (though its vowel was in an earlier period different) to met,
and so on. To describe such patterns is, in Saussurean terms, to describe
synchronically.

The prehistory of the phoneme

33

The first idea is in part far older than structuralism. The distinctiveness
of consonants, especially, was the principle behind the invention of the
alphabet; and in at least one other system of writing, devised for Korean
in the fifteenth century, features were represented too. But its current
formulation can again be traced back to the 1870s. The study of speechsounds was by then a separate discipline, called (in English since the
1840s) phonetics. Our story in this section is accordingly of one aspect
of that discipline.
Let us begin in Britain, with the work of the Anglicist and phonetician Henry Sweet. Sweet was among other things an acute observer of
phonetic detail, immortalised outside the history of linguistics in the character of Henry Higgins in George Bernard Shaws Pygmalion. But the
more exactly we discriminate between the sounds that speakers produce,
the more we tend to obscure the differences that matter to them. The ee
of heel, for example, is physically longer than the i of hill: that they are
different, certainly, any speaker will appreciate. But in compensation, as
Sweet noted at one point, the l(l) of hill is also longer. An l is also long
before a consonant such as d in build, but shorter before its counterpart
in built (Sweet, 1971 [1877]: 138). The duration of sounds is relative; and
for vowels, in particular, Sweet distinguished in principle five degrees of
quantity, though, for practical purposes, three were generally enough
(1971 [1877, 1890]: 137f.).
Such detail is compounded when we look at other languages. The l of
heel is, in modern terminology, alveolar: produced, that is, with the
tongue touching the flat ridge behind the upper teeth. But in French fil
the tip of the tongue is further forward, reaching the teeth themselves.
The l in French may also be regarded as front-modified, in comparison
with the deeper-sounding l in English: the tongue, as Sweet describes
it, is more convex than in English, its upper surface being arched up
towards the front position of , for example, y in yet (1971 [1908]: 113).
Much more could be said about the multiplicity of l s in either language.
But for speakers of each they represent a single consonant. They are
written alike (either l or, for quite unconnected reasons, ll ); and that
causes no confusion. Even when the l of French fil is pronounced with an
English accent, or the l of heel by a speaker whose accent is different
from the hearers, what will be heard is still the same l. The ee of heel also
varies, for example, between a Scottish and a Southern British accent.
But replace it with the vowel of hill and speakers will be aware that the
word is different.
The term speech-sound was, in effect, ambiguous. In one sense it
meant what, despite the numerous inherited deficiencies of writing systems,
the term letter had meant since antiquity. It was a unit distinguishing

34

Sound systems

one word or syllable from another, and a characteristic of each letter was
a sound value distinct from that of other letters. As David Abercrombie
showed many years ago, that sense of letter was in Sweets time not long
obsolete (Abercrombie, 1949).
But in another sense a speech-sound was, quite literally, a sound; and
the variety of sounds that could be differentiated was far greater than the
letters of any language. How then, in particular, were the sounds of
speech to be represented in writing? With all the nuances that a phonetician can hear, or just those aspects to which speakers are sensitive? The
answer depends, as Sweet saw, on our aims. One aim might be to record
on paper a specific accent, or the dialect of a small area. We must therefore note down everything that might be relevant in the fullest scientific
detail. Thus, in recording my accent, we must note not just that the vowel
in hill is shorter than in heel, but also that its quality is different: in the
notation now used, [h0l] not [hil]. But the quality of the l must also be
noted: it is what phoneticians call a dark l, articulated with the body of
the tongue raised towards the back of the mouth, as distinct from a clear
l in, for example, silly. Thus, again in later notation, [h0n]. But that is still
not all the detail we may need. We have not yet shown that the l is
longer; in other accents [0] may have a slightly different quality, and so
on. Such scientific representation was in Sweets terms Narrow; and,
following his usage, phoneticians now talk of a narrow transcription.
Another aim might be to develop a system of spelling. This could be
for a language not yet written; but, even for English, the dream of
Spelling Reform was not to fade until years later. For that, however, we
need to represent sounds in the other sense. We need to show that hill
and heel are not the same word; therefore, at some point, they must be
written differently. For example, if hill were written hil, heel might
instead be written hifl, with f showing that its vowel is longer. Alternatively, hill could be written h0l and heel hil, showing that the vowels
have different qualities. But in neither case would there be any need to
show more. Our aim is practical, to show as efficiently as possible which
word is which. We would defeat it if we were to try to represent the
sounds in scientific detail.
In Sweets terminology, such a system is Broad, indicating only broad
distinctions of sound (1971 [1877]: 231). A broad notation is one which
makes only the practically necessary distinctions of sound in each language and makes them in the simplest manner possible, omitting all that
is superfluous (1971 [1908]: 242). But what is practically necessary must
be what is distinctive; and that varies from one language to another.
In the term that Sweet used it is what is significant. Thus, in the earliest
of these references, he distinguishes the role in various languages of a

The prehistory of the phoneme

35

difference between narrow vowels (articulated with the tongue tense and
convex in cross-section) and wide vowels (with the tongue relaxed and
flattened). In French, the sounds in question are always narrow; hence
the distinction does not exist at all. In Danish or in Icelandic, it not only
exists but often makes the only difference between words. The distinction
is thus a significant one: that is, one that corresponds to real distinctions in the languages themselves. Therefore, in writing them, we have to
represent it. In Southern British English, the vowels of heel or pool are
narrow, those of hill or pull wide. But these are for Sweet distinguished
by their length; thus, though the distinction exists, it is not an independent one, being associated with quantity. Therefore, in writing English, its
representation would be superfluous. In this way, we may lay down as
a general rule that only those distinctions of sounds require to be symbolized in any one language which are independently significant ( [1877]: 230f.).
Sweets examples are from European languages, whose broad structure
was familiar. But suppose that we are investigating one that is entirely
unknown to us. We hear, for example, an l: to be precise, we hear a
sound that we perceive as l in our own language. So, in our notes, we
write l. At other times we hear, in the same sense, a d. Therefore we
write d . But, on reviewing our notes, we find that what is evidently the
same word has been written differently on varying occasions. Naturally
we check. We ask the speakers of the language to repeat what they have
said; and, yet again, we sometimes hear l and we sometimes hear d.
Sometimes, when we heard l the first time we will now hear d, and
sometimes the opposite. What has gone wrong?
Such difficulties were not new; but at the end of the nineteenth century
they were particularly serious in North America. The Amerindian languages were numerous and contact with them was in many cases recent.
They were, to put it Eurocentrically, exotic; still worse, they were very
different among themselves. The achievements of field linguists tend to be
unsung. But two, in particular, live in the folk memory of linguists. One
was Edward Sapir, who worked widely on a multitude of native American languages from 1910 onwards. The other was the anthropologist
Franz Boas, who had taught him. In 1900 Boas was already in his forties,
and saw clearly how our problem arises. As speakers of one language
we will hear sounds in another as we hear those of our own. Thus, as
speakers of English, we perceive an l when we hear a sound that is like
an English l, and so on. But our task as linguists is to work out what, in
Sweets term, is significant in the language we are studying. We must
instead hear what its own speakers hear.
One naive reaction had been to give up. The language was primitive,
and those who spoke it sloppy. Examples of American languages, Boas

36

Sound systems

said, have been brought forward to show that the accuracy of their
pronunciation is much less than that found in the languages of the civilized world (Boas, n.d. [1911]: 11). But that was not true. Every language, of whatever kind of society, has its own characteristic phonetic
system, distinguishing a definite and limited group of sounds (10f.).
Our problem is that the distinctions drawn by one are not those drawn
by others. Boas discusses, in particular, a sound in Pawnee (historically
of the Great Plains). It may be heard, he says, more or less distinctly
sometimes as an l, sometimes an r, sometimes as n, and again as d ; but in
the phonetic system of Pawnee it, without any doubt, is throughout the
same sound. As Boas describes it, it is an exceedingly weak r, made by
trilling with the tip of the tongue a little behind the roots of the incisors;
so, as soon as the trill is heard more strongly, as it might be in the
context of some neighbouring sound, we receive the impression of an r.
But it is a trill produced by the lateral part of the tongue adjoining the
tip, so that the centre of the tongue hardly leaves the palate: this makes
it at once potentially l-like, since in an l like that of English the side of the
tongue is distinctively lowered; but also, since its articulation is weak,
d-like. In addition, it is often accompanied by an audible breathing
through the nose; hence the further impression of an n, since that is distinctive for English n (11f.). Differences in sound that are significant in
our language are matters of mere detail in Pawnee; as, of course, vice versa.
Boas did not talk of phonemes; nor Sapir before the 1930s. But the
term sound was again potentially ambiguous. The sound in Pawnee is
described by Boas as throughout the same. In that respect it is like the
l of heel, hill, and so on. But at another level it is not the same: for example, it is often partly n-like but, by implication, sometimes not. In
that respect it is like the dark l of hill and the clear l of, again in my
accent, silly. Now in the first sense we are clearly talking of abstractions:
l in English, for example, is an abstraction which subsumes the range of
[n]s and [l]s and so on that, in a literal sense, one hears. What is the nature
of such abstractions?
The answer eventually given owes much to a theoretical analysis of
alternations, by the Polish linguist Jan Baudouin de Courtenay. In English,
the ee of meet is one sound (in phonetic notation [if] ); the e of met ( [t] )
is another. But they are related in that, here and elsewhere, they distinguish words whose meanings systematically contrast. Thus, in meaning
as in sound, meet is to met as lead is to led, as sleep is to slep- in slept,
and so on. We also find an [if] in adjectives like obscene and serene,
alternating with [t] in the nouns obscenity, serenity. Such alternations
cannot be explained by simple similarities among sounds. As mere sounds,
[if] and [t] are less alike than, for example, [if] and [0]. So too are, for

The prehistory of the phoneme

37

example, [k0] (as in drive) and [0] (as in driven). These too alternate: as
drive is to driven so ride is to ridden, as div[k0]ne is to div[0]nity so sal [k0]ne
is to sal[0]nity, and so on. But in no such pairs are [if] and [0] related
similarly.
Baudouin dealt with alternations in his own and other Slavic languages, and in a monograph in the 1890s he developed his theory in
remarkable formal detail. It was a theory of speech-sounds: again, in the
term then current, of phonetics. But phonetics, for Baudouin, had two
separate branches. In the first branch, sounds were studied simply as
sounds. As such they were physical events, distinguished auditorily by a
hearer and produced by the vocal organs of the speaker. In Baudouins
term, their study formed the anthropophonic branch of phonetics, or
anthropophonetics.
But sounds as such are transitory. In Saussurean terms and Saussure
knew of Baudouins ideas they belonged to what the Cours calls
parole. Therefore there had to be another branch of phonetics, to deal
with units in a speakers mind that correspond to them. It was these that
entered into alternations: thus, in the minds of speakers of English, the
[k0] of drive or divine is related to the [0] of driven or divinity, the [if] of
meet to the [t] of met, and so on. Their study formed a separate branch,
which Baudouin called psychophonetics. What were these units? In
Baudouins terminology, they were specifically the phonemes of the language. Thus the diphthong of drive was one phoneme (Phonem), the
[0] of driven another. But a unit at the psychological level was no longer
strictly a sound; it was a mental unit corresponding to sounds. The
phoneme, as Baudouin defined it, was a unitary representation in the
phonetic domain (eine einheitliche, der phonetischen Welt angehrende
Vorstellung), which arises in the mind through the mental amalgamation of the impressions received from the pronunciation of one and
the same sound (welche mittlelst psychischer Verschmelzung der durch
die Aussprache eines und desselben Lautes erhaltenen Eindrcke in der
Seele entsteht). It was the mental equivalent of the physical speechsound (= psychischer Aequivalent des Sprachlautes) (Baudouin de
Courtenay, 1895: 9).
As such, the phoneme is associated with a certain set of individual
anthropophonic representations (eine gewisse Summe einzelner anthropophonischer Vorstellungen), either articulatory representations, of
physiological tasks that can be executed, or the acoustic representations
corresponding to them. Thus, to take an earlier example, if l is a phoneme in English, the dark l of hill and clear l of silly might be said to
represent two different anthropophonic representations corresponding
to a single l in psychophonetics.

38

Sound systems

Baudouins theory was originally developed in Kazan, at the eastern

edge of our academic world, and it is not until the 1920s that, in work by
Sapir at its opposite extremity, we find hints that the abstraction might
be taken further. There is no evidence that Sapir had read Baudouin: nor,
in the beginning at least, Saussure. But in his introduction to language
he talks briefly of a system that lies behind the purely objective system
of sounds. The latter is already peculiar to a language, and corresponds
perhaps to Sweets broad representation. But back of this there is a
more restricted inner or ideal system, in which phonetic elements
are related in part independently of their objective character. This inner
system may indeed persist historically as a pattern . . . long after its phonetic content has changed. Every language, Sapir says, is characterized
as much by its ideal system of sounds and by the underlying phonetic
pattern as by its grammatical structure (Sapir, n.d. [1921]: 55f.).
These remarks were developed further in an article published four years
later. Sapir begins by comparing two sounds: one, not a speech-sound,
which would be made by someone blowing out a candle; the other a
sound objectively similar, made by many speakers of English (though
fewer now than when he was writing) in words such as when or wheel.
The most essential point of difference, he says, lies in the relation of wh
to other speech sounds. It is one of a definitely limited number of sounds
which belong together in a definite system of symbolically utilizable
counters. Each is characterised by a distinctive and slightly variable
articulation and by the corresponding acoustic image. But it is also
characterised and this is crucial by a psychological aloofness from all
the other members of the system. The relational gaps between the sounds
of a language are just as necessary to the psychological definition of these
sounds as the articulation and acoustic images which are customarily
used to define them. By contrast, the candle-blowing sound forms no
part of such a system (Sapir, 1949 [1925]: 35).
From this beginning, Sapir leaps to a conclusion that is potentially far
more radical. Within such a system, as can be seen already, each sound is
unconsciously felt as placed with reference to other sounds; if it were
not so, it would be no more a true element of speech than, for example,
a lifting of the foot is a dance step if it cannot be placed with reference to other movements that help to define the dance. But in a footnote
to the first use of placed Sapir explains that this has, of course, nothing
to do here with place of articulation . One may feel, for instance, that
sound A is to sound B as sound X is to sound Y without having the
remotest idea how and where any of them is produced. It is therefore
not enough simply to pattern off all speech sounds as such against
other sounds produced by the organs of speech . There is, in addition,

The prehistory of the phoneme

39

a second phase of patterning which is more elusive and of correspondingly greater significance for the linguist. This is the inner configuration
of the sound system of a language, the intuitive placing of the sounds
with reference to one another (35f.). In the remainder of the paper Sapir
shows by degrees how abstract he conceives it to be. The place that
sounds have in it is not affected by the different ways in which two
different speakers may produce them (36f.); nor by variations in the way
they are produced in contexts formed by other sounds (37f.). With these
cleared out of the way, we arrive at the genuine pattern of speech sounds.
But in the light of what he has said it almost goes without saying that
two languages, A and B, may have identical sounds but utterly distinct
phonetic patterns, or that languages with mutually incompatible phonetic systems, from the articulatory and acoustic standpoint, may nevertheless have identical or similar patterns (38).
The evidence which reveals the place of elements is partly that of
alternations. Thus, in English, the f of wife is related to the v of wives
just as, for example, the [s] of house, which like wife is singular, is related
to the [z] of plural hou[z]es. But we must also consider how units are
related in sequence. Thus, in English, p can be preceded by s: at the
beginning, for example, of a word like spoon or at the end in words like
cusp. So too can t and k: thus star and scum ( [skkm] ) or hoist and ask.
On that and other evidence these belong together in a coherent set
(42). Sounds that are physically alike may then, on similar evidence, not
belong together. In sound alone, the ng of sing ( [p] ) stands to a g ( [g] ) as
an m to a b or an n to a d: in phonetic terms g, b and d are plosive; ng,
m and n are nasal. But in spite of what the phoneticians tell us . . . no
nave English-speaking person can be made to feel in his bones that the
nasals belong together. In particular, both m and n can come at the
beginning of a word. An ng cannot, and in this respect it patterns as if it
were two elements, first a nasal and then a plosive, not one (43). The
whole aim and spirit of this paper, Sapir tells us in his final paragraph,
has been to show that phonetic phenomena are not physical phenomena
per se (45).
Let us try to sum up. We have seen that every language has its own
sound system, often, as in those that Boas and Sapir had studied, very
different from our own. Within it, a determinate set of speech-sounds are
distinguished by specific features, those that Sweet had called significant. What is significant in one language may, again, not be significant
in another. But such a sound may vary: sometimes longer, sometimes
shorter; sometimes with a range of qualities that, to phoneticians or to
speakers of another language, seem quite different. Hence the need, in
Sweets terms, for both broad and narrow representations.

40

Sound systems

Within the system, each sound is related to other sounds. It is related,

first, to those from which it is distinguished by significant features: English [if] from English [0]; English [g] (plosive) from English [p] (nasal).
Such relations might be said to define what Sapir called the objective
system of sounds. But they are also related by their patterning: in the
way they alternate or in the sequences they form. Such relations define
what Sapir called an inner system.
By the 1920s all this was, with hindsight, clear. But it is also clear
how such ideas connect with those of Saussure. For the sounds we are
describing have their existence only in the system of each language. The
Pawnee weak r is identified by the distinctions between it and other
sounds of Pawnee; the l in English by the distinctions between it and
other sounds of English. As mere sounds they are nothing, since their
physical properties are not constant. In a system of speech-sounds, as in
the Saussurean system of linguistic signs, there are no positive terms.
Readers may have wondered why I have devoted a whole section to ideas
that were initially so scattered. But it was in them that Saussurean theory
found a concrete application; and from their synthesis the movement that
we now call structuralism was born.
3.2

Phonology

The synthesis was above all that of N. S. Trubetzkoy. Trubetzkoy was

thirty in 1920, and had escaped from Russia to, eventually, a chair in
Vienna. Hence, among other things, the customary German spelling
of his name. With Roman Jakobson, who was another Russian migr
eventually in Brno in Czechoslovakia, he became a leading member of
what became known as the Prague School. He died in 1938, soon after
Austria was incorporated into another tyranny, with his great work on
The Principles of Phonology largely but not wholly finished. It was
published in the Travaux of the Prague Linguistic Circle, whose first
volume (1929) had included his first major contribution.
Let us begin with the term phonology. It had in origin a sense like
that of phonetics: thus, for Saussure, phonologie had been the study of
the physiology of speech-sounds, to be distinguished (in a usage subsequently peculiar to French-speaking linguists) from phontique as
the study of sound change (Saussure, 1972 [1916]: 55f.). But by the 1920s
a new distinction was needed. For by then, as we have seen, the properties of speech-sounds as sounds were one thing; their place in the
system another. In Baudouins terminology, the first was the domain of
anthropophonics; for Trubetzkoy and his associates, and in their wake
all later structuralists, the term phonetics was now restricted to it. The

The prehistory of the phoneme

Phonology

41

change was made by, among others, Baudouin himself (Baudouin de

Courtenay, 1972 [1927]: 280). Phonology was then the nearest equivalent to his psychophonetics.
Shifts in terminology are not in themselves interesting. But phonetics
and phonology could no longer be said to form a single science of speech
sounds (what had earlier been called phonetics). Phonology was concerned with specific languages. It was therefore part of what Saussure
had called the linguistics of language systems (la linguistique de la
langue), or linguistics in the strict sense. Phonetics was a discipline that
strictly lay outside linguistics, and had clear links with the wider study of
physiology or acoustics. From a linguists viewpoint from the viewpoint
of a practitioner of la linguistique de la langue it became ancillary.
In his Principles of Phonology, Trubetzkoy began by drawing a distinction like that in Saussures Cours (27ff.) between an act of speech
(Sprechakt) and the language or language structure (die Sprache oder
das Sprachgebilde) that a speaker and person spoken to must share.
Each has two sides which are related reciprocally: one signifying (die
bezeichnende) and the other signified (die bezeichnete). So, there must
first be a signifier in each individual act of speech. This lies in the audible
flow of sound, which will naturally include all manner of features with
which linguists, in the Saussurean sense, are not concerned. Thus, to
recall an illustration from the Cours, the form Messieurs! may be uttered
physically in grossly different ways; but in the structure of French it
remains, for any linguist, the same word. In each act of speech there is
also something signified, and that too will include much that is peculiar
to the individual context. But, in abstraction from each individual concrete message, we must then distinguish what is signified in the language
structure. This is not concrete, but is formed by rules which, in a specific
language, differentiate and order units of meaning. As elements of the
language structure these are finite in number. In parallel there must likewise, in the structure of the language, be a signifier. This is in turn
formed by the rules which, again in a specific language, differentiate and
order units of sound. In individual acts of speech, signifiers are formed
from an infinite variety of physical sounds. But, in the language structure, differences must again be finite (Trubetzkoy, 1939: 56).
Trubetzkoys account is clearly inspired by Saussure, and the overall
symmetry of his scheme is characteristic, as we will see in later chapters,
of this period. But in the case of signifiers, we must now distinguish
rigorously between two disciplines. One is concerned directly with the
physical sounds that form the signifying side of acts of speech. This was
henceforth what most linguists called phonetics, and is a natural science
dealing with the material aspect of speech-sounds (die Wissenschaft von

42

Sound systems

der materiellen Seite der (Laute der) menschlichen Rede, 14). It is precisely,
as a definition at the beginning of the decade had made clear, an auxiliary
discipline (discipline auxiliaire de la linguistique) whose subject matter
is the sounds of speech in general, in abstraction from their functions in
the language (traitant des phnomnes phoniques du langage, abstraction faite de leurs fonctions dans, if we may substitute the Saussurean
term, [la langue]) (Prague Linguistic Circle, 1931: 309).
Phonology, in contrast, is a discipline concerned with signifiers in
the language structure. It is the part of linguistics, therefore, dealing
with the phenomena of sound from the viewpoint of their functions in
the language (partie de la linguistique traitant des phnomnes phoniques
au point de vue de leurs fonctions dans la langue). This is again from
1931; and, as the Principles make clear, a phonologist considers as a
sound nothing that does not have such a function. Der Phonologe hat
am Laut nur dasjenige ins Auge zu fassen, was eine bestimmte Funktion
im Sprachgebilde erfllt (Trubetzkoy, 1939: 14).
The basic unit of this discipline is then, as for Baudouin, the phoneme. But for its new definition it will help if we look back to Sweet. For
what has a function in the language structure is whatever, in Sweets
term, is significant. Thus, in English, the l of hill is significantly different from, for example, the t of hit or the d of hid. It has a function in
English, in distinguishing one word from another; therefore it belongs to
an account of the phonology of English. But the l of hill is not significantly different from the l of heel. What is significant in these words
was, in Sweets analysis, the difference between a relatively short and a
relatively long vowel: in broad notation, [hil] versus [hifl]. Therefore
that alone has a function, and that alone has relevance, as Trubetzkoy
now put it, to the phonological system of the language.
The phoneme is then defined precisely by what is significant. In
Trubetzkoys terms, the dark l of hill would as a whole have one phonetic shape or sound form (Lautgebilde). The t of hit would have
another; so would, for example, the clear l of silly. As sounds, all three
are different. But a clear l and a dark l never distinguish one word
from another. Therefore the phonetic feature that differentiates their sound
forms is not relevant, in his term, to the phonological system. The features shared by both a dark l and a clear l do, however, distinguish
words with, in a broad notation, [l] from words with [t] or, for example,
[d]. Compare hill, hit and hid; silly and city (in a standard notation [s0li]
and [s0ti] ); Billy and Biddy. Therefore these features, and these alone,
do have relevance to the system. The [d]s of hid or dip, for example, are
distinguished by one relevant feature from the [t]s of hit and tip; by
another from the [l]s of hill or lip; by another from the [m]s of him or

The prehistory of the phoneme

Phonology

43

mill; by yet another from the [b]s of fib or bill. A phoneme, such as d,
can accordingly be defined by just the features that distinguish it, as one
unit, from all others. In Trubetzkoys own formulation, it can be said,
in general, to be the sum of all the features that are phonologically
relevant (die Gesamtheit der phonologisch relevanten Eigenschaften) in a
sound form (Trubetzkoy, 1939: 35).
There were other definitions of the phoneme, as we will see in a moment. But none shows more clearly how Sweets insights, as a practical
phonetician, could be elucidated by Saussurean theory. Each phoneme
is an element of a system in the Saussurean sense, defined by intersecting differences between it and other elements. Each difference was in
Trubetzkoys account an opposition, and each opposition (Gegensatz)
involves one or more phonetic features. For example, d in English is
distinguished from t by the role of voice (= vibration of the vocal cords)
in its articulation; from b by its articulation with the tongue against the
ridge behind the upper teeth (alveolar), and so on. The phoneme itself is
not a sound, identifiable independently. It is an abstraction, constituted
solely by the set of relevant or distinctive features, such as voiced or
alveolar, by which it is opposed to other similar abstractions.
The rest follows straightforwardly. As abstractions, phonemes enter
into other relations: in their role, as we have seen from earlier studies,
both in alternations and in forming sequences within words. Thus, among
the rules that form signifers in the structure of English is one by which,
in initial position, s may precede, for example, k (s[k]um, skill, and
so on), but not follow it. For Sapir such relations had established an
inner system, implicitly distinguished from the objective system of
sounds. For Trubetzkoy, they were similarly different from the oppositions by which phonemes were themselves defined. The term morphology had been used since the nineteenth century to refer to inflections
and other aspects of the shape (Greek morph8) of words. In studying
such relations we accordingly investigate the use, in morphology, of the
phonological units of a language. In the morphology of English there
are also alternations linking signifiers like drive, drove and driven. Such
relations had been central, as we have seen, to Baudouins psychophonetics. In Trubetzkoys account, both alternations and relations of
sequence within signifiers were a part of morphonology or, more fully,
morphophonology (Trubetzkoy, 1931: 160).
If morphonology was concerned with Sapirs inner system, the objective system of sounds, as he described it, was the topic of phonology
itself. But, in the light of what we have said, this wording is no longer as
appropriate as it may at first have seemed. For, crucially, a phoneme is
not simply a sound. It is instead an abstraction from a multitude of

44

Sound systems

sounds, and is characterised solely by particular features that are constant. We must therefore envisage a further relation by which, in the
terminology of Trubetzkoy and of most other structuralists, it is realised
by or manifested by the sounds from which it is abstracted. A speaker
of English who utters, on a particular occasion, the word hill can be
expected to produce at the end a sound, [n], which we can identify as a
typical realisation, in that position, of the l phoneme. But this sound
that we identify is not itself that phoneme. A speaker who utters the word
leave will normally produce at the beginning a clear [l] which we
will identify as a different realisation of l. But l itself is no more an
[l]-sound than an [n]-sound. It is no more than the set of features by which
one symbolically utilizable counter, again in Sapirs words, is distinguished in the structure of the language from all other counters. The
sounds by which it is realised are describable only as the concrete variants, as Trubetzkoy defines them, of a unit of the language structure that
is itself invariant (Trubetzkoy, 1939: 36).
These points are dealt with rather briefly. But their importance in the
Saussurean tradition will at once be obvious. In the analogy that has
caught the imagination of so many readers of Saussures Cours, the
express train from Geneva to Paris is an abstraction realised, on differing occasions, by a different engine, driver, set of carriages, and so on.
Despite this it is on each occasion the same train, by virtue simply of
the differences between it and all others in the timetable. But analogies,
however seductive, cannot be entirely convincing. The development of
phonology showed for the first time, through the analysis of language
itself, how a deliberate abstraction from the physical reality of speech can
advance our understanding in a way that is impossible if the physical
phenomena are studied directly. To describe sounds as mere sounds was
to embark on a sea of infinite detail. To describe them as the realisations
of phonemes was to reveal a finite underlying order that, to those who
discovered it, was exactly of the kind that Saussure had foreshadowed.
A final problem was that of method. The origins of phonology had
lain, in part, in practical linguistics: in studies of languages or dialects in
the field, and in applied phonetics. Both traditions continued, the latter
in the work, especially, of the English phonetician Daniel Jones. The
theory offered a foundation, therefore, for techniques that were tried and
useful. Could these now be codified?
How, in particular, was a linguist to work out what the phonemes of a
language are? Let us imagine, for the sake of argument, that the structure
of English is unknown. We encounter people who speak it, and our first
task is to discover which distinctions of sound are significant. The way
we will do so followed naturally from the theory that has been explained.

The prehistory of the phoneme

Phonology

45

We will begin by looking for word-like units that have different meanings. For example, we find that hill is different from hit, full from foot,
and so on. Let us say that such words contrast. We then look for minimal contrasts, between units with the shortest time span. For example,
we find that foot contrasts with feet; feet in turn with leat; feet with feel.
Failing evidence of still shorter time spans, we thus form the hypotheses
that [f ] at the beginning of a word contrasts minimally with [l], [t] at the
end of a word with [n], and, between them, [v] with [if]. In this way, we
establish contrasts of sound in specific positions. Then, as a final step, we
relate sounds that are found in different positions. Thus, at the beginning
of words, we have distinguished an [l] and, at the end, an [n]. We also find
sounds that resemble both in medial position: for example, in silly and
fuller. But nowhere will we find words in which one l is in minimal
contrast with another l. Therefore we conclude that they are variants of
a single phoneme, which we distinguish in all positions from all other
phonemes.
The method was developed in elaborate and rigorous detail in the
1940s, in an early paper by the American Charles F. Hockett (Hockett,
1942). But the basic rules had been set out by Trubetzkoy (1958 [1935]:
10ff.; 1939: 41ff.) and have been taught in practice to most apprentice
linguists, in one version or another, ever since. They have thus become
banal; and it is hard for any writer of a later generation to recapture the
enthusiasm that this unity of theory and method once inspired. Decades
later, one American linguist who was young in the 1930s was to describe
the phoneme, not entirely in jest, as his darling (Hill, 1980: 75). Its
attraction was, in large part, that the theory and method were so closely
linked, each new and each informing the other.
But the link was so close that, in America in particular, the method
came to dominate. In particular, the definition of the phoneme, which in
Trubetzkoys account had flowed from basic principles or assumptions,
came to rest instead on codified procedures by which, in an ideal application, phonemes could be induced from data.
Let us return once more to the l in English. From a phoneticians
viewpoint, it is effectively a range of variant sounds: those that, in
Trubetzkoys term, realise it. Thus, for Jones, a phoneme was a family
of sounds. It was a family in a given language, of sounds related in
character and used in such a way that no one member occurs in a word
in the same phonetic context as any other member. Thus a dark [n] is
related in its phonetic character to a clear [l], and there is no phonetic
context, as defined by a position in the word or by specific sounds preceding or following, in which both are found. I have cited the definition from
a book first published in 1950 (Jones, 1962 [1950]: 10). But Jones was

46

Sound systems

then nearing seventy, and such definitions had been in use since about
1916 in practical language teaching (vii). A formulation similar in
essentials had been given earlier in his practical account of English phonetics (Jones, 1975 [1918]: 49).
Definitions of the phoneme have been a topos for commentators. For
many phonologists it was in the beginning still a unit of what Baudouin
had called psychophonetics. It had a psychological content and reflected
a psychological distinction between different kinds of phonetic opposition (Trubetzkoy, 1929 [Vachek, 1964: 109]; 1936 [Vachek, 1964: 188] ).
Phonetics, roughly speaking, was concerned with what one actually pronounces; phonology with what one thinks one pronounces (ce quon
simagine prononcer) (Trubetzkoy, 1933: 232). In the same vein Sapir
had written in the 1920s of the psychological aloofness of one counter
in the system from another. But by the end of the 1930s this way of
talking had become redundant. We take for granted that a community
of speakers share a language system or a language structure. This has
the same effect as Bloomfields fundamental assumption (2.2), of establishing the study of such systems as autonomous. The phoneme is one
element in their structure, and can be defined by the distinctive oppositions that each phoneme bears to others. If our assumptions are valid
there is no need to refer specifically to anything outside our discipline.
Where a Saussurean assumed that there is a language system, a
Bloomfieldian assumed, as we have seen in 2.2, that there are units that
recur in utterances. Such units were distinguished by sound features, and
a phoneme, as Bloomfield defined it, was a minimum unit of distinctive
sound-feature (Bloomfield, 1935 [1933]: 79). A phoneme too is thus a
unit that recurs in utterances.
Suppose then, for the sake of illustration, that I say The little bottle is
still full. The phoneme l recurs, in this sense, five times: at the beginning
and the end of little, and so on. Its precise phonetic quality varies, even
in positions that are similar, in -ittle and -ottle. What recurs is thus
again a unit of . . . sound-feature, not literally a sound. But by the same
token it can also be seen, as Jones had seen it, as a family of sounds;
and it is as such that, in practice, the next generation in America went on
to define it. For the method that had been developed has in essence two
steps. The first is one of segmentation. Thus, in this and other utterances
in English, we identify the shortest sounds that contrast: in that way we
segment The little . . . into [2] followed by [i] (the) followed by a clear
[l], and so on. The second step is one of classification. We class together,
into a single family in Joness sense, the clear [l] at the beginning of
little, the dark [n] of full and the similar post-consonantal sound in
bottle, with (in my speech) the slightly clearer sounds of still, following

The prehistory of the phoneme

Phonology

47

[0] instead of [v], and -ittle. All that remained, then, was to define this
family formally by operations on our data. To that end, the first need
was for watertight criteria which, when applied to the analysis of utterances, would in principle identify contrasting segments. From the end of
the 1930s segments were, in a term that was usual in America, phones.
Every phone was then an allophone of a phoneme. Hence there was
a further need for watertight criteria which, when applied to sequences
of contrasting segments, would in principle group different phones as
allophones either of the same phoneme or of different phonemes.
The in principle is important. The object was no longer to develop
practical methods of analysis, but to set out the ideal procedures by
which phonemes might be operationally defined. But it is clear, from
Joness definition in particular, that a phoneme as a class would have to
meet two main criteria. One was that its members never contrast. In the
terminology used in America in this period, allophones of a phoneme
had to be in complementary distribution: the distribution of one,
defined by the positions in which it is found or of the sounds preceding
or following it, must not overlap that of another. There were problems
in making this criterion watertight, exceptions had to be taken care of,
and so on. But in an account which was to pass into textbooks in the
1950s, the phoneme was defined, in part, as a class of allophones in
complementary distribution.
The other main criterion was that its members must be alike. In Joness
terms they were related in character; or, as the American tradition put
it, they would have to meet a test of phonetic similarity. Again there
were problems and, in particular, a requirement that all variants of a
phoneme should share a distinctive set of phonetic features, which was
implicit in Trubetzkoys account in the 1930s, could not always be sustained. But if, once more, the criteria by which phones are established
were watertight, and these and other principles of classification were made
watertight in turn, the phoneme needed no other definition. It was precisely the unit that, in the analysis of a body of data in any language,
would result from the rigorous application of the criteria which had been
set out.
We will return to phonology in the next chapter, where we must consider its implications, in the 1950s especially, for historical linguistics.
From then on its development, however fascinating, is that of one branch
of linguistics, and need detain us only in passing. But in its formative
period phonology was the heart of structural linguistics. In the 1870s
Sweet had believed already that phonetics, in the sense in which that
term was used in his day, was the indispensable foundation of all study
of language: be that study, he went on, purely theoretical or practical

48

Sound systems

as well. This, he said, was generally admitted ( [1877], cited with similar
passages in Sweet, 1971: vii). In the early twentieth century many of the
best minds in linguistics worked in this field, and it was in the theories of
Trubetzkoy and his contemporaries that what we now see as the main
ideas of Saussures Cours were first applied fruitfully. In later chapters we
will see how these ideas were also extended to other aspects of language.
But, again and again, when faced with problems of theory or method in
another domain, linguists were to turn to current phonological theories,
to concepts such as that of distinctive oppositions, or to the practical
methods of analysis that phonologists had developed, both for general
inspiration and for specific models.
The study of sound systems has in consequence remained a popular
and influential subject. The editor of Language, which is the leading general linguistic journal, reported in 1996 that in that year, of the hundred
or so typescripts sent to him, more than a quarter were in phonology as
it has later come to be understood ( [Aronoff ], 1997: 466). The figures for
1995 and later years, though lower, were still high. To observers in other
disciplines, the fascination of so many linguists with this aspect of language may at first seem puzzling. Systems of sounds are, after all, but one
quite small part of the total language structure. But it derives from a
tradition of research and teaching that was the centre of linguistics
throughout one crucial and exciting phase.
3.3

Structuralism

By the end of the 1930s Bloomfields newer trend of linguistic study, as

he had called it in the early 1920s, was formally christened. It is not
known exactly when or by whom the terms structural and structuralism were coined. But Trubetzkoy referred in German to the language
or the language structure (die Sprache oder das Sprachgebilde) (again
Trubetzkoy, 1939), and this usage can be traced back in the Prague School
for ten years. For the Danish linguist Viggo Brndal, structuralism was
by then already known by that name (Brndal, 1943 [1939]: 95). It seems
that the term clicked at once for everyone.
Its sense is made clear in the same year, in an introduction to phonology by the Dutch linguist N. van Wijk. A language, as he explains in the
preface, is a collective possession of a community (een collectief bezit
van een gemeenschap). This clearly follows Saussure. In the mind of
each member of such a community the language exists as a system with
a determinate structure (leeft de taal als een system van een bepaalde
structuur), and the task of a linguist is to work out such structures. The
history of a language is also necessarily the history of its structure; not

The prehistory of the phoneme

Structuralism

49

that simply of the individual words and other units, in abstraction from
their mutual relations (Van Wijk, 1939: xiii). Phonology itself is accordingly described in his title as a chapter of structural linguistics (een
hoofdstuk uit de structurele taalwetenschap). In an article of the same
year, H. J. Pos distinguishes the structuralism of Trubetzkoy and other
phonologists from the nominalism, as he describes it, that had informed
earlier descriptions of speech-sounds. In the nominalist view, such sounds
were no more than noises, studied individually without reference to the
mind of the speaker (sujet parlant) who produces them. Structuralism,
in contrast, has both placed them in the larger system to which they
belong and made clear that their reality lies in their relation to the consciousness (conscience) of individuals who speak and understand what
is spoken (Pos, 1939: 724). Pos was writing in Trubetzkoys memorial
volume in the Travaux of the Prague Linguistic Circle, and phonology is
central to his argument. But similar insights would be fruitful in the
study of meaningful units (74ff.).
It is perhaps a coincidence that Van Wijk and Pos were both writing
in Holland. But it is no accident that both took their inspiration from
phonology. For Jakobson, commenting on Van Wijks book, this was
not only the first part of structural linguistics to be realised; the phoneme,
as its fundamental concept, had thereby become a touchstone of structuralism generally (und gerade deshalb fiel dem Phonem als einem phonologischen Grundbegriff die Rolle eines Prfsteins des Strukturalismus
zu) (Jakobson, 1962 [1939]: 284).
Its extension beyond phonology belongs more to the 1940s and 1950s.
But let us return, by way of introduction, to the principle of contrast.
In comparing hill with, for example, hit and hid we establish contrasts
between (in a broad notation) [l], [t] and [d]. In the same way, if we
compare Walk up the hill with Walk down the hill and Walk over the hill,
we establish contrasts between up, down and over. In both cases, we
are performing operations in which something is held constant and
other things are changed. Thus, holding constant Walk the hill, we
replace up, in the position marked by a dash, with down or over. Likewise, holding [h0] constant, we replace [l], in the position following, with
[t] or [d]. When the replacements are made the meanings of the whole are
altered. Therefore the results in one case show that [l], [t] and [d] must
realise, or be members of, three different phonemes. In the other they
establish that there are differences, of some sort, between up, down and
over. Of precisely what sort we have not said: we will return to that
problem later. But the method of comparison is the same.
If we hold [0l] constant, we can replace [h] in initial position ( [0l])
with [p] ( pill ) or with [s] (sill ); if we hold up the hill constant we can

50

Sound systems

replace walk with, for example, run or climb; and so on. We thus establish
a structure in which units are related on two axes. On the one hand, they
are related to others that precede or follow. For example, there is a
relation in hill between [h] and [0]; likewise between [p] and [0] in pill, [h]
and [k] in hull and any other elements that can individually replace either.
There is also a relation between walk, run or climb and up, down or over.
This kind of relation had been described, in Saussures Cours, as
syntagmatic (see again Saussure, 1972 [1916]: 172f.).
On the other hand, each unit is related to the others that can replace it.
In the initial position in, for example, [0l], [h] is related to [p], [s], [f ],
and so on. In the second position in Walk the hill, up is similarly
related to other units such as down and over. Relations on this axis are
said, from this period onwards, to be paradigmatic. Our basic method,
therefore, is to take a stretch of speech in which we assume that units are
related syntagmatically, and to look for contrasts in specific positions.
We do not know at first how many positions will be established. Thus we
can only guess, ahead of our analysis, that [h0l] realises three successive
phonemes; or that Walk up the hill is made up of walk, up, the and hill.
But, by our analysis, we will validate relations on both axes. In establishing that [h], [0] and [l] stand independently in paradigmatic relations we
confirm that just these segments are related syntagmatically. In establishing that up, for example, is related paradigmatically to units such as down
and over, we also establish that it is itself a unit, which is therefore related
syntagmatically to, as we will establish similarly, walk. In failing to establish paradigmatic relations between segments like up with a shorter time
span (u-, say, or -p) we confirm that it is a minimal unit. When we get
down to basics, the method that distinguishes phonemes and the one by
which we validate larger units like up seem to be similar.
The term paradigmatic was first introduced by the Danish scholar
Louis Hjelmslev, in a conference paper in 1936 (Hjelmslev, 1938: 140). In
the Cours Saussure had called relations of this kind associative (thus,
for an example involving sounds, Cours, 180), and many commentators,
following Hjelmslev himself, have implied that associative had the same
sense. But Saussures concept was in reality much wider. A prominent
diagram (175) shows four series of associative relations in which the word
enseignement (teaching) is linked first to, among others, enseigner to
teach and enseignons we teach; then to other nouns whose meaning is
similar (apprentisage training, ducation education); then to others that
are formed with -ment (changement change, armement armament); then
to words of other kinds which end with the same syllable (the adjective
clment clement or the adverb justement justly, precisely). In short,
Saussure meant what might be expected from the term associative. The

The prehistory of the phoneme

Structuralism

51

relation is in general between any units that may be associated in the

minds of speakers.
In proposing the new term Hjelmslev said that his motive was to avoid
the psychologisme (n. 3) of the old. But by the end of the 1930s structural linguistics was not only autonomous. It was a technical discipline
concerned with problems of method and definition that Saussure could
scarcely have formulated. They had been solved, as we have seen, for
phonemes, and the next step was to solve them for other units. Of the
relations that Saussure had originally called associative, only those that
formed the axis of contrast were relevant.
The American school did not in general use this terminology. But it
was in America, in the 1940s, that the basic operation was most rigorously developed. In Walk up the hill, a grammarian will also see a close
relation between the and hill, forming a larger unit called a phrase. But
why do the and hill form a phrase and not, say, up and the? The issue was
explicitly addressed by Zellig Harris, in work first published in 1946.
In Walk up the hill we can replace the hill with it, with Ben Nevis, with
the mountain over there, and with many other such forms. We can also
replace it, with the same forms, when it is in other positions: for example, in The hill is too steep. There are no replacements, on the same
scale and applying so generally, for up the. For replacement Harris talks
of substitution. The method by which we analyse a sentence will accordingly require . . . no other operation than substitution, repeated time and
again (Harris, 1946: 161).

52

Diachrony

Diachrony

From the 1940s onwards structural linguists were increasingly distinct

from linguists in general. Structuralist schools were not established equally
in all countries: scarcely at all, for example, in Italy or in Germany. But
in the United States especially to be a linguist was increasingly to be a
structuralist. Above all, it was to be a descriptive linguist. The term
descriptive has many resonances, and can easily be misunderstood. But
for Americans its sense was like that of Saussures synchronic. In a
leading introduction to linguistics, descriptive or synchronic linguistics
was the study of how a language works at a given time, regardless of its
past history or future destiny (Hockett, 1958: 303). It deals with the
design of the language of some community at a given time (321).
But Saussurean linguistics also had a diachronic or historical branch.
How was that affected?
When the Cours appeared this branch was dominant. Saussure
himself had been young in Leipzig in the 1870s, when the so-called
Junggrammatiker or neogrammarians formulated basic principles of
change in language. It was through their work above all that, we are told,
linguistics in the true sense (la linguistique proprement dite) had been
born (Saussure, 1972 [1916]: 18f.). It is not surprising therefore that the
chapters of the Cours devoted to its diachronic side (Part 3, 193ff.) say so
little that was new. Of the structuralists who followed, Bloomfield had
been trained in the tradition of the neogrammarians, and his masterly
account of language change, which forms most of the latter half of Language, was based firmly on their principles. They were applied above
all in his brilliant work on the Algonquian family in North America,
summarised in one of his last papers (Bloomfield, 1946).
But the foundations of linguistics were by then seen differently. For
Hermann Paul, in the 1880s, the only scientific treatment of language
had been historical. The moment one goes beyond the mere registration
of details, the moment one tries to grasp the way they are related and
to make sense of the phenomena, one is at once in the field of history,
even if perhaps one is not clear about it (Sobald man ber das blosse
52

Diachronic phonology
Diachrony

53

Konstatieren von Einzelheiten hinausgeht, sobald man versucht den

Zusammenhang zu erfassen, die Erscheinungen zu begreifen, so betritt
man auch den geschichtlichen Boden, wenn auch vielleicht ohne sich klar
darber zu sein) (Paul, 1920 [1880]: 20). For the Danish scholar Otto
Jespersen, the historical view had brought about a vast change in the
science of language. Instead of looking at such a language as Latin as
one fixed point, and then fixing another, like French, in one classical
form, nineteenth-century linguistics viewed both as being in constant
flux, as growing, as moving, as continually changing. It cried aloud like
Heraclitus Pnta re . . . (Jespersen, 1922: 32f.).
But if one followed Saussure, Latin, for example, had indeed to be a
fixed point. It may be that, in the maxim of Heraclitus, all things are in
reality in flux. But, to study the history of a language, we must first
abstract successive stages through which, like the pieces in his image of
the chess board, it has passed. Each stage can be described synchronically,
and each description can stand, as a treatment of one state of the language, on its own. A diachronic account is secondary.
Bloomfields definition of a language was not, as we have seen, Saussures (2.2). But for him too all historical study of language is based upon
the comparison of two or more sets of descriptive data. To describe a
language one needs no historical knowledge whatever; if indeed we have
such knowledge and we allow it to influence our description, it is bound
to distort the data (Bloomfield, 1935 [1933]: 19f.). For a later Bloomfieldian, linguistics in a narrow sense could be divided into descriptive linguistics, which was concerned with descriptive grammars, and contrastive
linguistics, in which different descriptive grammars were compared.
Historical linguistics was a branch of contrastive linguistics (Trager, 1949).
This new view gave rise to new insights and new problems. The earliest
concern the history of sound systems: we will therefore take that field
first (4.1). But for a more general orientation it will help if we return to
Saussures parallel with a chess game (2.1). Each player, once again,
moves one piece at a time; similarly changes in a language were individual, and each led from an old state of the language (tat de la
langue) to a new one. Between moves the pieces form a determinate
pattern on the board. Similarly, a language was at any moment in one
state of equilibrium, which the next change would replace with a new
equilibrium. The only important difference, we are told, is that moves in
chess are made intentionally. A language, in contrast, has no intention
(la langue ne prmdite rien). Changes occur spontaneously and by
chance (spontanment et fortuitement) (Cours, 125ff.).
But is that always so? We must grant, of course, that no supervisory
agent alters languages deliberately. But speakers are agents, and changes

54

Diachrony

in a language system must originate in and be sanctioned by their speech.

Is it likely, therefore, that all changes are fortuitous? In a game of chess,
the players calculate their moves with reference to the present and future
position of the pieces. A queen is threatened and for that reason it may
be moved or the piece that threatens it may be taken. Now, once again,
a language does not calculate; nor do its speakers as a body change it
consciously. But might some changes come about not blindly but, in part
at least, as a response to the state its system is then in?
Let us take a famous example, based on studies contemporary with the
Cours, by the dialectologist Jules Gilliron. From the findings of the
French Linguistic Atlas, published by Gilliron and his assistant Edmont,
it was clear that in Gascony, from the Garonne southwards, a consonant
that was ll in Latin had changed, at the end of a word, to t. For example,
Latin coll(um) neck had become cot. It was also clear that, in the west
of this region in particular, the word for cock was an innovation. In
surrounding dialects it was derived from Latin gall(us); so, in the area
of this sound change, we would expect gat. But in the region around
Bordeaux the word for cock was instead related to French vicaire
curate; further south, to French faisan pheasant. It seemed, then, that
these dialects had undergone two separate changes. By a change in their
sound systems Latin -ll was now -t. By change in their vocabulary, the
expected word for cock had been replaced.
But in Gillirons account these changes were connected. For, in the
same dialects, the word for a cat (Latin catt(us)) had the form gat; and,
by the sound change, an earlier gall (cock) would also, as we have seen,
become gat. In rural communities this would have led to misunderstandings, which speakers could avoid only by replacing at least one of these
forms. Now alongside gat cat there was a feminine: gata female cat. Its
form was simple and transparent (masculine gat + feminine ending -a).
Alongside the word for cock there was also a feminine (hen), but its
form was less transparent. In Latin it had been gallina: gall(us) + -ina.
But ll did not change to t between vowels; instead it changed to r: garina.
The relation of gat cock to garina hen would thus have been obscure,
while that of gat cat to gata female cat remained clear. Hence, it is
argued, gat cat was retained, while gat in the sense of cock came to be
avoided. Speakers then found new ways to refer to cocks.
Gilliron was no structuralist and his general ideas must be inferred
from detailed analyses of these and other findings. But in Saussurean
terms the first sound change must be seen as one move on the chess
board. By it the expected word for cock would become, in the new state
of the language, a homonym of gat cat. But the language would then be
dysfunctional, since speakers need distinct forms with these meanings. In

Diachronic phonology

55

a metaphor that runs through Gillirons own writings, the dialects affected
are or would be in a partly pathological state (un tat pathologique).
Therefore therapy was needed; and this too was in part determined by
the state the system was then in. For by another move on the chess board
an original -ll- between vowels became or had become r; and, in the state
of the langue that that gave rise to, the forms for cock and hen would
be at best obscurely related. The therapeutic adjustment was made at a
point where the relations within the system were weak. The replacement
of gat (cock) is not, in this account, spontaneous, and all that was
fortuitous, in the next move, was the particular form (related to vicaire
or to faisan) that groups of speakers had recourse to.
Gilliron died in the 1920s and was, to repeat, no structuralist. But the
image of pathology and therapy, or the general way of thinking that goes
with it, was to inspire many others.
4.1

Diachronic phonology

If each language has a determinate structure, its history, as Van Wijk

remarked at the end of the 1930s, is the history of that structure (3.3). It
was in phonology that the synchronic study of this structure was first
fully realised. It is not surprising therefore that it was in phonology, in
the work of Van Wijk and his contemporaries, that a structuralist theory
of diachrony was first proposed.
One early essay is by Roman Jakobson. In the traditional account, as
he describes it, every sound is treated in isolation. Thus, in our example
from Gascony, the consonant derived from earlier -ll would have been at
one time like that of surrounding dialects; its articulation then changed,
in a way that we can treat with reference to this consonant alone, to that
of -t. But for phonologists a speech-sound is not isolated: each phonological fact is treated as a partial whole related to other partial sets
at higher levels (Jakobson, 1949 [1931]: 316). The first principle of historical phonology is, correspondingly, that each change must be treated
in relation to the system within which it takes place (toute modification
doit tre traite en fonction du systme lintrieur duquel elle a lieu).
Thus, in the example from Gascony, the change of -ll to -t would have to
be considered with reference to the relations, within the system, that this
unit bore to other units.
The nature of a phonological system was by then becoming clear
(3.2). Its elements were phonemes, each characterised, in the account of
Trubetzkoy and other members of the Prague School, by a set of distinctive features. In my form of English, for example, the vowel 0 (in pit)
shares one feature with, among others, t (in pet) and a (in pat). In a

56

Diachrony

phoneticians classification these are all front vowels, produced with the
front part of the tongue nearest the roof of the mouth. It is also distinctively close (produced with the lower jaw raised and the tongue near the
roof of the mouth). This feature it shares with, among others, the back
vowel v (in put). Among the front vowels 0 is then distinguished from
mid t (with the jaw more open) and open a (with full opening). As a
back vowel v can be distinguished in parallel from a mid m (in pot)
and an open k (in putt). This part of the system is thus formed by a
network of relations in which 0 ( pit) is to v ( put) as t ( pet) is to m
( pot) and a ( pat) is to k ( putt); this is shown diagrammatically in Fig. 1.
0

Figure 1

Suppose then that in such a system a distinction between phonemes

is lost. For example, every speaker of this form of English once distinguished vi (in moor or poor) from hf (in more or pour). Like v and
m these were respectively close back and mid back, and opposed as in
Fig. 2 to close front 0i (in pier) and mid front tf (in pear).
0i

vi

tf

hf

Figure 2

But many speakers now no longer distinguish them. In Jakobsons terminology (1949 [1931]: 319), a former distinction is becoming dephonologised. But this does not merely reduce the number of phonemes. It
also alters the system. Where there was once symmetry, with a pair of
front vowels balanced by an equal pair of back vowels, there would now
be asymmetry, with two front vowels, 0i and tf, but just one back
vowel corresponding to them.
In other instances a new distinction is phonologised (321). But that,
in turn, does not mean simply that the number of units is increased: we
must again ask how the system at large is affected. Now it is obvious
that, in either case, some changes will impair its symmetry: thus, for
speakers who have lost the contrast between vi and hf, front vowels

Diachronic phonology

57

are no longer balanced equally by back vowels. But others will enhance
it. Consonant systems, for example, often distinguish series of consonants
with the features labial, dental/alveolar, velar. These are produced
respectively with the lips, as English p; with the tip of the tongue behind
the upper teeth, as English t; and with the body of the tongue raised to
the roof of the mouth, as English k. Let us suppose then that at a
certain stage a language has three labials (p, b and m) and three
dental/alveolars (t, d and n). As p is to t and b is to d so m,
which is distinctively nasal, is to the corresponding nasal n. But there
might be no distinct nasal in the velar series: just k and g with no p.
So, as shown in Fig. 3, there is what is often called a gap or hole in the
system.
p

Figure 3

By a later change the gap might then be filled and the asymmetry
eliminated. It may seem tendentious to use words like impair and enhance. But might it not be that, all else being equal, systems change
in ways that will reduce asymmetry? In Gillirons metaphor, a lack of
balance is pathological. A change which leads to greater balance will be
therapeutic. It will bring the system closer to an ideal healthy state.
These are Gillirons terms, not Jakobsons; and, for the most part,
Jakobson did little more than set out logically the ways in which a change
within a system might be classified. But in the final section of his paper he
insists that when we look at a linguistic change in its synchronic context
it is within the scope of teleology. (Quand nous considrons une mutation
linguistique dans le contexte de la synchronie linguistique, nous lintroduisons
dans la sphre des problmes tlologiques (334).) This implies again that
changes are not blind and fortuitous; they have a purpose which, we are
then told, may be obvious. Thus a specific change might be preceded by
a disturbance in the balance of a system (une rupture de lquilibre du
systme). The change might then result in a removal of the imbalance
(une suppression du dsquilibre). If so, Jakobson said, we have no
difficulty in discovering (aucune peine dcouvrir) its function. Its
task is to restore the balance (sa tche est de rtablir lquilibre), which
had earlier been lost.

58

Diachrony

These brief remarks will not perhaps bear too much commentary. But
if we take this view the obvious question is whose purpose? There is
again no agent that watches over language systems and ensures that an
imbalance is corrected. So what precisely did this mean?
One thing it might mean is that phonological systems are selfregulating. Let us compare, for example, a thermostat. It is set to maintain
a stable temperature: so, if the temperature falls below that level, the
heating will switch on. The purpose of this adjustment, if we must speak
teleologically, is to restore the correct temperature. If the temperature
then rises too high, the heating will switch off. The aim is still to keep
the temperature steady. This analogy will not have been so obvious in the
1930s. But, similarly, for each part of a language system there are properties that will, in general, be maintained or maximised. In the case of
phonology these are, or include, those of balance. But just as in a building there are many factors, both internal and external, that continually
affect the temperature, so there are many factors that, as they impinge on
phonological systems, can lead to imbalance. A system will continually
initiate adjustments whose purpose, if we must again speak in that way,
is to restore the balance.
If this were so, we would have no need to speak teleologically. The
system simply conforms to laws that govern it. To the extent that they
are satisfied its structure is in balance. Where they are not it is unbalanced, and there is pressure to correct it. But new imbalances will
continually arise. Sometimes they will be brought about by changes that
originate outside the system. But they may also result from changes that
the system has itself made in response to earlier pressures. Thus, in correcting a disequilibrium at one point, it may create a new disequilibrium
at another. In Jakobsons words, this often leads to a whole series of
stabilising changes (toute une chane de mutations stabilisatrices).
This is certainly one interpretation of what Jakobson said. But if we
take it literally the pressures, as we have called them, are exerted directly on the system. The Saussurean langue, of which the phonological
system is a part, is seen as monitoring and adjusting its own state, independently of the community of speakers, who at any time will simply
instantiate it, in acts of parole, as it stands. The obvious objection is that
languages do not exist as independent entities. There would be no chess
game if there were no players. There would be no language system if there
were no speakers; and it is surely on them that the pressure truly bears.
The pressures implied were made clear in the early 1950s, in a series
of brilliant studies, written at first in exile in New York, by the French
linguist Andr Martinet. At the most general level speakers need, above
all, to be understood; therefore they are under pressure to maintain
distinctions between phonemes and, in this way, between words and other

Diachronic phonology

59

units of meaning. But, as Martinet saw it, they are also constrained to
spend as little energy, in speech as in other things, as they can. Linguistic
evolution is accordingly governed by the continual paradox (lantinomie
permanente) that, on the one hand, people have a need to communicate
and express their feelings (des besoins communicatifs et expressifs de
lhomme) and, on the other hand, they tend to make do with the minimum
of physical and mental effort (de sa tendance rduire au minimum
son activit mentale et physique) (Martinet, 1955: 94). Such pressures
are permanently opposed and never wholly reconciled. The more clearly
speakers try to distinguish between one sound and another, the greater
the effort they must make. But, if they make too little effort, the distinction will be at risk.
We may therefore expect sound changes to reflect both pressures. Sometimes, when in practice there is no risk of misunderstanding, a distinction
will no longer be kept up. Thus, in my form of English, there was no
serious danger, in most contexts where these words are used, that poor
might be mistaken for pour, moor for more, and so on. But other changes
tend to enhance distinctions. In the history of Spanish, for example, a
consonant that was historically a [ts] had changed, by the early seventeenth century, to [r]: thus [katsar] to hunt (spelled cazar) became, as in
present-day Castilian, [karar]. In the same period another, that was historically like [q] in English ship, changed to [x]: for example, [bjeqo] old
(viejo) became, as generally in modern Spanish, [bjexo]. Before these
changes, the relations into which these phonemes entered could be displayed as in Fig. 4
p

ts

tq

k
g

Figure 4

(compare Martinet, 1955 [19512]: 323). Their effect was to replace it

with the system in Fig. 5,
p

Figure 5

tq

k
g

60

Diachrony

in which, as Martinet points out, the consonants are more clearly differentiated. In its former state, both [ts] and [q] were phonetically similar
to other sibilant (or s-like) units: a [tq] (ch), as in modern fecha date,
and an [s], as in casa house, which is produced in modern Castilian with
the extreme tip of the tongue against the teeth-ridge. In Martinets words,
the system suffered from a concentration of sounds in this region (ce
systme souffre de trop de concentration dans le domaine des sifflantes),
while other articulatory possibilities, which were plainly open in a language with relatively few phonemes, were not exploited (323). In the new
system, the phonemes are more clearly distinguished.
In cases like this we can see how a system might be represented as, in
effect, correcting an imbalance. Thus, in terms of the features that these
tables imply, the change of [ts] to [r] also fills a gap in the t series and
that of [q] to [x] a corresponding gap in the k series. But the locus of
change lies in the habits and perceptions of successive groups of speakers.
To distinguish the consonants in viejo or cazar some would have shifted
their articulation, at first slightly and at first sporadically. But the system
could, at that stage, still be said to be the one they had inherited. Such
shifts were soon familiar; they would have become less slight and more
frequent; and finally new generations, as they learned to speak from the
example of their elders, would have interpreted what they commonly
heard as [r] or [x] as indeed, in terms of their distinctive features, r and
x. Their language had, we can then say, the new system. An important
factor, in Martinets general theory, is the attraction of an integrated
system (lattraction exerce par le systme intgr) (Martinet, 1955: 80).
Thus, in the case of gaps or cases vides (80ff.), the system in general, in
which series of phonemes stand in parallel relations, exerts a pull on
sounds that are not fully integrated. But the pull is exerted on a changing population of speakers, and in particular, as in Martinets discussion
of a later change in Spanish (84f.), on new speakers.
In the same way systems can be seen as influenced by other systems.
Les langues, as Martinet remarks at one point, do not evolve in an
ivory tower (89). People who speak one language or dialect will engage
with people who speak others; many are bilingual. Nor are the boundaries of a speech community established for all time. A population may
speak language A; but then, through contact with a population that
speaks language B, they may more and more speak that instead. Later
generations may lose A altogether, as, for example, many descendants of
Welsh speakers now speak only English. We all know that, in learning a
new language, we will speak it with a foreign accent. Habits of pronunciation will be carried over from the one we speak already. So, when
language A gives way to language B, we might expect that habits of this
kind might sometimes be retained. For example, aspects of the English

Diachronic
System
and phonology
norm

61

spoken in Wales or Ireland might reflect specific features carried over

from the Celtic languages.
When Martinet was writing this had long been familiar. But here too a
structuralist will look at systems and not simply at sounds. In mediaeval
Spanish, for example, there had also been a phonological distinction
between voiceless and voiced sibilants: voiceless ts, tq, s and q were
thus opposed to voiced dz, dx, z and x. But over the centuries Castilian in the north gained many speakers from Basque; and in contemporary Basque, as Martinet reconstructs it (1955 [19512]: 317), voiced
sibilants were not separate phonemes. Let us imagine, following Martinet, a community of Basque speakers who learn Spanish as a foreign
language. They would hear a word in Spanish with, for example, a ts;
this they would identify, as a phoneme, with the ts in their own system.
But they would also hear words with a Spanish dz and, since [ts] and
[dz] did not realise different phonemes in Basque, they would identify
that too as realising their ts. They would accordingly pronounce both
consonants alike; and, in practice, native Spanish speakers, though themselves distinguishing ts and dz, might not usually misunderstand or
need to correct them. Their descendants then progressively speak Spanish
more than Basque, and in time Spanish only. But the dialect of these
descendants, on the northern boundary of the larger area in which Spanish
is spoken, would be one in which pairs such as ts and dz were no longer
distinguished. Now it was from these speakers that, as the use of Spanish
continued to spread, still further generations of Basque speakers would
learn it. That dialect of Spanish would thus become more widespread. It
was also from the very north, as Martinet reminds us in another section
of his study, that the Christian Reconquista started. For speakers to the
south this dialect then became a model to imitate (310), and, for other
Spanish speakers, the distinctions were in time lost.
The explanation for these changes is external. Martinet was invoking
the impact, on the system of Spanish, of another system. For those cited
earlier, by which [ts] became [r] and [q] became [x], it is internal. We
were invoking an inherent imbalance. In each case, when we speak in
that way, we are dealing with an abstraction. Spanish, qua langue, was
not directly under pressure, at one stage, to adapt to Basque any more
than, later, it was directly under pressure to enhance its symmetry. But
the mapping of one system onto the other is essential to a structuralist
understanding of what happened.
4.2

System and norm

Let us return, in the light of these insights, to Saussures analogy of the

chess game. It was now in one way more exact, in that not all change in

62

Diachrony

language is, in the words of the Cours, spontaneous and fortuitous.

But in another way it was too simple. To describe a move in chess, we
need say only that piece A was on one square and is now on another
square. Nothing more is relevant. Likewise, we might say, a consonant
that was x in an earlier state of Spanish became, in a subsequent state,
q and, in a still later state, x. But these changes were in practice
neither instantaneous nor separate; both spread gradually, over many years,
through a large community. It is inconceivable, as Martinet remarks,
that the whole nation, from Burgos to Granada should have pronounced
a word like viejo, in which j had earlier been [x], in one way in 1550, and
in a new way, with [x], in 1625 (1955 [19512]: 319f.). What do we
suppose, he asks, of a speaker whose lifetime spans this period? As the
changes spread the usage of successive speakers will have varied, with
their age, with the region in which they lived, and with their social status.
For it is also clear that the ruling classes were the last to give up the old
forms (320).
Let us take another, somewhat simplified illustration, this time from
the order of units in sentences. In modern English one does not say, for
example, Then rode Alfred to Winchester. The verb must instead follow
the subject (Then Alfred rode to Winchester). But the order of Then rode . . .
is normal in related Germanic languages and was once normal in English
also. Therefore, in Saussurean terms, the system has changed: what was
once required is now excluded. But we can hardly imagine that at a
certain moment every speaker switched from the inherited order to the new.
For a time both would have been heard; and, at first, the new arrangement
might have been rare. Then, gradually, it began to prevail: fewer speakers
used the old one, they used it less often, and the range of words with
which they used it narrowed. Eventually the modern order was so normal
that, as children learned the language, they would hear the old one only
in a range of residual uses. Therefore they would learn it only as an
exception; and, for later generations, its scope was set to narrow further.
In both cases we must envisage a gradual shift in usage: in which, at
first, more Spanish speakers normally said [x] and, by the end, more
normally said [q] or [x]; in which the majority of English speakers may,
in one period, have preferred to say Then rode Alfred . . . and, in a later
period, to say Then Alfred rode . . . It is further evident that, in phonetics
at least, there can be shifts of usage that will leave the system unchanged.
In an article in the 1930s, Archibald Hill had remarked that, in the Southern United States, the whole . . . vowel system would strike a phonetician coming from New England as a notch higher (= closer) than in
other parts of the country. These differences must have their origin in
earlier changes, by which usage in one population diverged from that

Diachronic
System
and phonology
norm

63

of others. But between the dialects there were no striking differences in

phonemic pattern (Hill, 1936: 15). Therefore, in Hills terms, the historical shifts had been phonetic, not phonemic. In the terms that Jakobson
had used a few years earlier, there was no change to the phonological
system: no difference between sounds was either phonologised or dephonologised, and the features that distinguished phonemes were also
unaffected. All that had shifted were the precise vowel qualities by which
the system of oppositions was realised.
The problem for Saussurean theory is clear and is central. For, in the
view that had been presented in the Cours, each language is a system
of invariants; and it is to such a langue or language system, which exists
in each community as a fait social or social reality, that an individual
speaker has to conform. Change in a speakers language is thus change
in the system. But at the level of speech a change will not be instantaneous, either across a community or for each individual in it. There will
instead be periods in which usage varies from one form of realisation to
another; and, even where it is claimed that the system does not change,
similar patterns of variation may be found. These patterns too are part
of the social reality to which speakers conform. An American born and
bred in Charleston or Atlanta will be socially constrained to speak like
others in the Southern states, a Bostonian like others in Boston. A grandee
born in Spain in the 1540s would have spoken throughout his life like
others of his age and social standing; a peasant born in 1600 like others
of his district. Each individuals language, in an ordinary sense, will
conform in such ways to the usage of the community in which they are
at home. But there is a discrepancy between a language in that sense,
and the processes of change in such a language, and the Saussurean
notion of a system, and of change as a transition from one system to the
next.
The most constructive critique was that of the Romanian Eugenio
Coseriu, in work first published in Montevideo in the early 1950s. For
Saussure, the term parole had referred to concrete acts of speaking:
to individual acts of utterance by individual speakers in individual
circumstances. A langue was, in contrast, at once abstract and a social
construct. There was in this way an initial identification of what was
individual with what was concrete (parole) and what was social with
what was, as Coseriu put it, formal or functional (Coseriu, 1962 [1952]:
53). But we can see now that Saussures dichotomy had conflated things
that are logically separate. A distinction, for example, between phonemes
is indeed both functional and social. The way in which a phoneme
will be realised is, however, social but not functional: it does not have
relevance, again in Trubetzkoys sense, to the phonological system. But

64

Diachrony

it is not simply individual. In realising l in English, speakers do not

freely articulate a clear [l] or a dark [n] as each individual is inclined.
Instead they follow a set of norms by which it is clear or clearer in some
positions in a word and dark or darker in others. The term norm is
Coserius, and norms too are social.
Coserius solution was to recognise not one but two successive levels
of abstraction. At a higher level we can establish a system. But it is no
more than a system of possibilities (es sistema de posibilidades), of
coordinates specifying what is open to speakers and what is excluded
(que indican caminos abiertos y caminos cerrados) (98). Our example
from syntax has no parallel in Coserius own illustrations. But, in a simplified story, English as a system may at one time have allowed both
orders. In this construction one could, in principle, say Then rode Alfred
. . . ; and one could also, in principle, say Then Alfred rode . . .
But the system is then not the primary abstraction from speech. What
in reality is imposed on an individual (que . . . se impone al individuo)
is instead conformity to a norm. Let us take, from among Coserius
illustrations, the words in Spanish for actress (actriz) and for manageress (directora). One is formed from actor actor with -iz and the other
from director with -a; and, in general, the system of the language, as a
system of possibilities, allows feminines to be formed with either of
these endings. Hence the word for manageress could, in principle, be
directriz and the word for actress could, in principle, be actora. Now
any speaker, naturally, will learn from other members of the community
to say actriz and directora. But in using these forms, they are constrained
not by the system, which would equally allow these meanings to be carried by actora and directriz, but by what Coseriu calls the normal realisation of the system. Only at this level of realisation, as he puts it, are
the forms that speakers actually learn and use preferred (79). In our
example from syntax, it is possible that, for a period in which the system
allowed either order, different speakers were constrained by norms that
varied. Conservative speakers may have usually realised such constructions, though perhaps not always, with the verb before the subject. Perhaps, in particular, they put the verb first when a word like modern then
preceded. Others may have usually realised it, though perhaps again not
always, with the verb after the subject.
The norm which constrains a speaker is thus, in effect, a system of
prescribed realisations (un sistema de realizaciones obligadas), of
social and cultural requirements (de imposiciones sociales y culturales)
(98), at a level of abstraction (95f.) between speech, as a representation
of individual concrete acts of utterance, and the system in a Saussurean
sense. The preferences it imposes are, of course, not absolute. For the

Diachronic
System
and phonology
norm

65

essence of creativity in language, as Coseriu explains in various places,

lies in exploiting the possibilities that the system will leave open in a way
that goes beyond the limits that the norm prescribes. Furthermore norms
vary according to the community (98).
Coserius account was based on a critique of fundamental concepts,
without specific reference to diachrony. But, among its many applications, it was claimed to be especially revealing for our understanding of
linguistic change (106f.). Some changes, we can now say, affect only the
norm. In Southern British English, the vowel in words like hair, which
earlier phoneticians represented with a diphthong ( [ti] ), is now more
usually a monophthong ( [tf] ). But its place within the system, in relation
to the vowels of hear, hire and so on, has not altered. Some Spanish
speakers might create new feminines in -ora: senadora (female senator)
is one that is not in my dictionary. If such a word became part of the
norm, the language, in an ordinary sense, would have a new form. But
its creation would have followed a route that in the system, as Coseriu
describes it, was already open.
In other cases changes will affect the system also. It is the norm, as
we have seen, that is directly imposed on individuals. But every member
of a speech community has knowledge of the system (tiene conciencia
del sistema); and, in speaking, will remain within the limits that it sets
([se mantiene] dentro de las posibilidades del sistema). Individual speakers
then, in addition, either know or do not know the norm and, in their
speech, either obey or do not obey it. Within the system, as we have
said, there is scope for originality in expression; and what is at first original in a single speaker (the use of senadora, for example, or a sporadic
monophthongisation of [ti] ) may be taken as a model by other speakers.
In consequence, what was at first an isolated error or an isolated act of
creation may in time become part of a new norm, to which each individual will in turn conform and sometimes not conform. But norms
cannot change without affecting what Coseriu calls the balance of the
system (el equilibrio del sistema). At any specific moment the norm
reflects this balance. So, as the norm changes, the balance of the system
is in turn altered, to the point at which, in the end, it overturns completely to one side or another (hasta volcarse totalmente de un lado o de
otro) (107). In this way, Coseriu argued, changes that originate at the
lowest level of abstraction, in the speech behaviour of specific individuals,
can successively change both the norm, which is an abstraction at an intermediate level, and the system, at a level still higher.
Our example from syntax is again not modelled on one that Coseriu
gave. But, in these terms, we could speak of a moment of overturning
when, as the norm had shifted more and more to the order of Then Alfred

66

Diachrony

rode . . . , and the balance of the system had in turn altered, it finally
became the only order that, exceptions apart, the language allowed. In
the history of Southern British English, to return to an earlier example,
the normal realisations of the vowels in moor and more have shifted
gradually from [vi] and [hf], which were once distinguished clearly by all
speakers, to variants that are increasingly convergent. This will at first
have been a matter of realisation only. But in time the system may allow
no opposition.
Neither the monograph cited (Coseriu, 1962 [1952] ) nor a later important book on change in language (Coseriu, 1973 [1958] ) had the attention,
or the critical review, that they deserved. But the role of variation was to
catch the imagination of many linguists in the 1960s, through the work of
the American sociolinguist William Labov. For Labov argued both that
variant realisations could themselves form structures, and that variation
could be striking even in quite tight communities.
It will be obvious at once that, in any community, the speech of any
member will differ slightly from that of any other. This is in part, but
only in part, because they are physically different. Hence, for example,
Coseriu had spoken of an individual norm as well as a social norm
(Coseriu, 1962 [1952]: 96f.). But Labov claimed that, if it was studied in
isolation, the speech of an individual might make no sense on its own.
The dialect of an individual is often called an idiolect; and in the city
of New York, where his classic study was carried out, most idiolects do
not, in phonology at least, form a coherent system. They are studded,
he goes on, with oscillations and contradictions, both in the organization of sounds into phonemes, and the organization of phonemes into
larger systems (Labov, 1966: 6). The same speakers might on one occasion have an r in their pronunciation of a word like car or card; on other
occasions they might not. The quality of the vowel in word a might
sometimes be recorded as identical to one recorded in word b; at other
times as different. We cannot, of course, recover the facts from past
stages of a language. But through Labovs and later studies it became
clear that, as sounds change, the speech of individuals may be inconsistent. In our example from Spanish, x changed, in an overall perspective,
to q and then x. But, as the changes spread, the realisation of the
consonant in a word like viejo may at any time have ranged, for a certain
group of speakers, between [x] and [q]; for another, between [q] and [x];
for another, over all three.
The most coherent system, in New York at least, was that which
includes the . . . community as a whole (7). But a system, for Labov as
for Saussure, was simply a set of differences. Within the New York
system, the vowel of, for example, bad is different from, or is opposed to,

Diachronic
System
and phonology
norm

67

that of bed. But how these vowels are realised is another matter. In fact,
among the speakers Labov studied, the phonetic quality of a was found
to vary strikingly according to, in particular, the social class to which
he assigned them and the style, as he described it, in which they were
speaking. The lower a speakers class the more its realisation was towards one end of a scale; the higher their class the more it was towards
the other. For speakers of any given class, the more casual their speech the
greater would be the percentage of lower variants; the more careful, the
greater that of higher variants. To put this in Coserius terms, we could
still speak of a norm for each individual. It might be represented by a
frequency distribution, correlated with styles. But the distribution would
reflect a larger pattern which is systematic in the community. Only when
the whole community is studied can its order be revealed.
Labov was a pupil of Uriel Weinreich, whose own thesis, on the effects
of contact between different language systems, had been supervised by
Martinet ( Weinreich, 1963 [1953]: x). It is perhaps not surprising, therefore, that he was particularly interested (318) in cases where there was
evidence of change. Let us suppose that, as historically in the example
of the Spanish consonants, a sound change is spreading through a community. As Spanish changed, we would have expected the distribution of
variants to be different, at any stage, in different places and at different
levels in society. But at any moment there would also have been differences among speakers of different ages. The older the speaker the more
often, at least, we might have expected to hear [x] or [q]; the younger the
more often at least, all other factors being equal, [x].
Such findings are, as Labov describes them, in apparent time; and
from them we can conjecture, even without evidence of earlier stages,
that a change is happening in real time (319f.). In New York r was
variably realised after vowels in words, as we have said, like car or card.
It was found, among other things, to be less frequent in casual style and
among speakers lower in class. But its frequency also varied with the
age of speakers; and this confirmed what was indeed known in advance
of Labovs investigation, that it had spread quite recently to this area. In
Labovs terms, this was a change that was still in progress; also one
that, as was attested by the distribution across classes and styles, had
spread from above. The details of this case need not concern us (see
Labov, 1966: 342ff.). What is important is the overwhelming evidence
that, in phonology at least, the process of change is reflected in the
patterning of variation. Before his death in 1967, Weinreich had begun
a major paper on the theory of change in language; in its completion,
which reflected Weinreichs thinking, Labov laid down the principle,
among others, that all change involves variability and heterogeneity. It

68

Diachrony

is thus transmitted within the community as a whole; not confined to

discrete steps between parents and children. Linguistic and social factors are again closely interrelated in it (Weinreich, Labov and Herzog,
1968: 188).
But, for the community as a whole, Labov still spoke of a system. In
New York, for example, the vowel of bad was still different, in the
Saussurean sense, from that of bed. Their realisation might vary systematically, but both entered into a system of differences constant across
the New York speech-community. By the end of the 1960s the terminology of Saussure was increasingly supplanted by that of Chomsky, which
we will come to later. A system was accordingly a set of rules, and a
linguistic variable was then an element within the system controlled by
such a rule (Weinreich, Labov and Herzog, 1968: 167). The quality of a
vowel was one such variable; the presence or absence of an element, such
as r in car or card, another. But the system as a whole, including the
variables, would again be the same for every member of the speech community whose language it represents.
A famous illustration was that of copula deletion in northern American Black English. A speaker of this variety of English could in principle say, for example, He is a fool, with a form of the copular verb to
be, or He a fool, without it. Therefore the system, for all speakers, allows
both. In Labovs account, the set of rules includes one for the deletion
of is. But it was not a categorical rule: one that speakers have to follow
always. It was a variable rule, controlling a variable (is present or is
absent) which is a function of all the factors that determine the frequency
of one value or the other (Labov, 1977 [1969]: 93ff.; compare Weinreich,
Labov and Herzog, 1968: 16972).
The system of Black English must then be different from those of
other varieties of English, by which He a fool is excluded. It would also
change, as a system, if in the future He a fool were to become the only
possible form. In our own example from syntax, the original system would
have had to change at one stage to allow the order of elements in Then
Alfred rode . . . In Labovs terms we could claim that there was then a
variable rule by which the order was rearranged. Though the facts can
at best be surmised, the frequency with which it was followed would have
been explained by various factors, and their influence would again have
changed successively from one period to another. At a later stage we
might claim that the rule itself changed: while the rearrangement had at
first been from Then rode Alfred . . . to Then Alfred rode . . . , now it was
the opposite. There would then be a new system, though the rule, at that
stage at least, was still variable. In such accounts, linguistic change is in
reality continuous: it is not just a matter of discrete moves on a Saussurean

Diachronic phonology
Universals

69

chess board. But at a more abstract level the transitions from one state
of the language to another are still there.
4.3

Universals

The theory that had been developed in the 1950s was one in which change
in a language system, towards, for example, greater symmetry of vowels or
consonants, originates in changes in the behaviour of individual speakers.
It is on individuals that, in particular, pressures operate. Where sound
change was concerned this theory had been taken, by 1960, as far as then
seemed possible.
But words and grammar also change and here too change can be transparently of systems. In Latin, for example, hic (translated this) was
basically opposed as a first person pronoun (this in the sense of proximity to or association with the speaker) to both second person iste
( proximity to or association with whoever the speaker is addressing) and
third person ille (proximity to or association with neither). But the languages that have descended from Latin have in general a two-way distinction similar to that in English: thus, for Italian, questo this, quello
that. In the history of Italian, it is not the forms alone that have changed.
The system of differences has also been transformed.
The example is one explored in the 1940s by the Romanist Walter von
Wartburg (Wartburg, 1969 [1943]: 208ff.). But here and elsewhere Wartburg tends to talk as if a language system governed itself. At a certain
stage the language can try to maintain a distinction (la langue peut
tenter de maintenir la distinction) by looking elsewhere for a new
element to supply it (en cherchant ailleurs un remplacement). Alternatively, it can buy peace (elle peut acheter sa tranquillit) at the price
of sacrificing it (au prix de la renonciation toute diffrentiation) (209).
A language can again be said to have avoided the weakening of a distinction (. . . a obvi cet affaiblissement . . .) (210). This way of talking
seems once more to suggest that language systems are self-regulating.
Like, for example, thermostats (4.1), they monitor and adjust their own
states. Alternatively, such talk is figurative; but a figure for what, precisely?
The image of self-regulation has been taken from an essay on structuralism in general, by the Swiss psychologist Jean Piaget (Piaget, 1968:
13ff.); and, though rarely made explicit, it has remained, and remains,
tempting. It is not surprising, therefore, that the issue was raised again
in the 1970s, by theories that proposed laws regulating patterns of word
order.
The origin of this idea lies in the early 1960s, in work in America by
Joseph H. Greenberg. It was purely descriptive, and his findings were not

70

Diachrony

absolute. But let us assume that, in any language, we can identify the
elements subject, verb and object. In some, they are normally in that
order: thus, in English, subject Mary plus verb saw plus object Jim. But
in others their order is different. Let us also assume, for example, that
we can identify a category of what are described as adpositions: either
prepositions, like English to, which precede a noun (to Cambridge) or
postpositions, with a similar role but coming after it. We can then ask if
there is a correlation between the order of adpositions (preposition versus
postposition) and that of, in particular, verbs and objects. In a sample of
thirty languages examined by Greenberg, some had as their dominant
order VSO (verb first, then subject, then object). Such languages, he
found, were always prepositional (Greenberg, 1963: 78). This correlation
was accordingly described as a linguistic universal, or universal of language. Others had the normal order SOV (first subject, then object,
then verb); such languages were found, with overwhelmingly greater than
chance frequency, to have postpositions. This too was a universal (79).
In all, Greenberg postulated forty-five such universals, some absolute
and some not, involving these and other syntactic elements.
Let us assume, for the sake of argument, that the assumptions underlying this study were valid. Thus, in particular, all languages have, in
an equivalent sense, subjects and objects. But why, we then ask, should
there be such correlations? A tempting answer is that universals reflect
laws that govern the evolution of language systems. Suppose that a language has the normal or dominant order SOV; but, unlike the majority
with that order, it has prepositions. By law, we claim, it should have
postpositions. Therefore it might be expected to change in conformity
with the universal. Perhaps it too might, in time, become postpositional.
Alternatively, the ordering of S, O and V might change: perhaps to that
of English (SVO). In either case, the change is explained (we would
claim) by pressure to conform to a law that every language system must
ideally obey.
This line of thought was developed in particular by Winfred P. Lehmann
and by Theo Vennemann. In Lehmanns account, the order of prepositions or postpositions was one of several features that covary with, specifically, the order of the verb and object. Another of particular importance
is the position of an attributive adjective. In French, for example, blanche
white is attributive in une maison blanche a white house; and the noun
to which it is related (maison house) comes first. This is the order we
expect, from Lehmanns theory, if, as again in French, the verb precedes
the object. A language was then said to be consistent if such expectations are met. So, in these terms, French is a consistent VO language
insofar as (a) the object normally follows the verb, (b) it has prepositions

Diachronic phonology
Universals

71

instead of postpositions, and (c) the position of most adjectives is after

the noun. A language was inconsistent to the extent that expectations
are not met. Thus English is, in one respect at least, an inconsistent VO
language. Once more (a) the object normally follows the verb; (b) it has
prepositions; but (c) the normal position of adjectives, as in a white house,
is before the noun and not after.
Various arguments followed. For example, from the aberrant position of adjectives in English we may propose the hypothesis that it is a
language which has been changing from OV structure, in which we would
expect an adjective to come before the noun, to VO structure (Lehmann,
1978: 37). But what if the order in English were to change, so that this
aberrancy is eliminated? In the same volume Lehmann speaks of a
drift in languages towards consistency (408). Studies of syntactic change
show that, as patterns of the new type . . . gradually become established,
the language becomes increasingly consistent (409). In change, languages
of a given type are modified in accordance with the principles that he
proposes (431); such principles permit explanations. So, if the position
of the adjective in English were to change, that might be explained by
pressure for consistency to be enhanced.
The term drift had been used much earlier by Sapir (n.d. [1921]:
150, 154ff.). But in what exactly does the current on which a language
drifts, or the pressure by which it is impelled in that direction, consist?
In Lehmanns account, the basic patterns of any language are governed
by simple but powerful principles (396). In, for example, I saw a white
house, the attributive adjective (white) comes between the verb (saw)
and the object noun (house). But by a principle that he formulates
(19), the sequence of verb and object noun must not be interrupted.
If English were to change, the language would be responding to that
principle.
In Vennemanns account, the role of white in relation to house was that
of an operator to an operand. So too was that of an object in relation
to a verb; also, among others, that of a noun in relation to a preposition
or a postposition. He therefore proposed a general principle, by which
languages tend to serialize such elements unidirectionally. In English,
objects are among the operators that must follow verbs: so, taking X as
any operator of this sort, it is a VX language. In tune with this, an
operator such as Cambridge in to Cambridge follows a preposition; if the
tendency were perfect, attributive adjectives should also follow nouns.
In XV languages the pattern will be the opposite. Operators such as
objects come before verbs; adpositions like in are instead postpositions;
attributive adjectives precede nouns. The history of the word order syntax of each language can be understood to a large extent, according to

72

Diachrony

Vennemann, as a development toward consistent implementation of this

principle (Vennemann, 1975: 288).
The principle is said to be of natural serialization. The development
of languages is accordingly, all other factors being equal, towards greater
naturalness. But where again is the pressure that will propel or guide
them in that direction? Those who speak a language such as English do
not themselves find its inconsistency unnatural. There seems no reason
why, in a construction like I saw a white house, the noun should not
continue to come after the adjective, simply because that is the pattern
that each speaker will learn. How could the application of this principle
lead, in appropriate circumstances, to change?
In the paper cited, Vennemann considers in particular the motives for
a change from XV order to VX. There is a tendency, he claims, for
languages to be of the first type (XV) when the subject and object are
reliably distinguished by case endings: for example, in Latin by a nominative ending (servu-s slave or puella girl) and an accusative (servu-m
or puella-m). Where there are no such endings, as in English, they tend to
be of the other (VX). Now that as it stands is no more than a finding,
generalising one of Greenbergs (288f.). But suppose that a case system
breaks down, so that subjects and objects are no longer distinguished
effectively in that way. This can happen, for example, through sound
changes that affect the endings. If the language is, as we expect, an XV
language it then changes, according to Vennemann, to VX (289). But
why, he asked, should that change follow?
His answer (289ff.) was that, when the cases are lost, the roles that
nouns play can be obscured. In English, for example, it is clear that, in
the man who Mary saw, the subject of saw must be Mary. It is clear
because English is an SVX language; hence, if Mary were the object,
the order would be the man who saw Mary. But what if a language has
the order SXV? So long as it retains its cases there is no difficulty. If the
word for Mary is the subject it will be nominative and the word for
who accusative: thus, using English words for clarity, who-accusative
Mary-nominative saw. If the word for Mary is the object, it will be the
opposite: thus, with XV order, who-nominative Mary-accusative saw.
What then if the cases are lost? If Mary is the subject it will again precede
saw: who Mary (S) saw (V). But its position will be no different when it is
the object: who Mary (X) saw (V). Sometimes, when such phrases were
uttered, hearers would still understand, in practice, what the speaker
meant. But quite often they would not.
This is one of several ambiguities and perceptual difficulties (290) to
which Vennemann points. An XV language without cases is, in brief,
dysfunctional. Its speakers will accordingly find other ways to make their

Diachronic phonology
Universals

73

meaning clear; and, in Gillirons terms, apply therapy to it. But in

explaining such changes Vennemann had not invoked a specific law of
naturalness. Speakers are simply under pressure to make clear what they
mean. Hence, as Martinet argued, they are under pressure to preserve
distinctions between phonemes (4.1). Hence too they are under pressure
to preserve distinctions in syntax. In phonology this may lead to changes,
for example, in the structure of a consonant system. In syntax it may
lead, for example, to changes of word order.
What was the status, then, of principles like that of natural serialisation? On one view they are still no more than summary findings. We
find that there are correlations between one construction and another:
typically, in Greenbergs formula, with overwhelmingly greater than
chance frequency. This is just a matter of observation. To explain it we
must look much deeper, in the need for speakers to make clear what they
are saying, for their hearers to understand them with as little effort as
possible, and so on. Likewise when we look for explanations of change.
We might find that, all else to the best of our knowledge being equal,
changes are in the direction of greater consistency. That would again be
no more than a finding. We cannot explain it by hypostatising laws that
govern systems independently of their speakers. To propose such principles is merely to restate the initial findings. We must again look deeper,
at the pressures bearing on speakers.
But that was not apparently what Lehmann, in particular, was claiming. Just as Wartburg, for example, could talk of a language finding
ways to avoid the weakening of a contrast, so a language, on a literal
reading of what Lehmann said, is under pressure to maximise consistency. Was that just a convenient way of talking? Or were languages to
be seen as, in reality, autonomous systems, subject to constraints that,
if no countervailing factors intervene, require consistency?

74

The architecture of a language system

The architecture of a language system

Let us return to synchronic linguistics. By the end of the 1930s, the description of sound systems had a theoretical foundation in the work of
Trubetzkoy and others, and its method was increasingly codified. But the
phonology of a language is again just one part of the total language
structure. How do phonemes relate to, in particular, units of meaning?
By what methods can these in turn be identified securely? What kinds of
relation does each unit of meaning bear, in turn, to other units of meaning?
These were technical questions, and the elaboration of techniques of
description, which is the most striking feature of linguistics in the decades
that follow the Second World War, was not at first accompanied by new
general or philosophical ideas. For the most part European linguists tended
to found their work on those of Saussure, and linguists in America on
Bloomfields. The most exciting problems were of method, and those that
concerned the detailed structure of a language system. For Trubetzkoy
the language structure had already been not one system but several
partial systems (mehrer Teilsysteme) (Trubetzkoy, 1939: 6). These components hold together, complement each other, and stand in mutual
relations (so da alle Teile einander zusammenhalten, einander ergnzen,
sich aufeinander beziehen). The main task for his successors was to say
exactly how they do so, and exactly what the components are. It was in
America especially that answers were given, and much of what was worked
out in this period has found its way into textbooks and been taught for
decades to generations of students. It is therefore a large part of what
structural linguistics has concretely achieved.
We must begin, however, with the theory of Louis Hjelmslev, who was
more nearly Trubetzkoys contemporary. Its roots lay in the 1930s, and
at the time it was felt to be so original that a new name, glossematics,
was coined for it. With other Europeans of this period, Hjelmslev found
his inspiration in the Cours de linguistique gnrale, and it was by him
that the overall design of language systems, seen as systems of linguistic
signs as Saussure had conceived them, was most rigorously and systematically worked out.
74

Expression and content

5.1

75

Expression and content

The definitive account was published in the war years, in a monograph

On the foundations of linguistic theory (Hjelmslev, 1943). It was never
easy, and in the 1950s was already the subject of a secondary exposition longer than the original (Siertsema, 1965 [1955] ). But at the heart
of Hjelmslevs theory lay an analysis of the sign relation, or what he
called the sign function, that linked the two sides of a sign. For Saussure,
as we have seen, each sign had been a unit with two faces: on the one
hand, what the Cours had called an acoustic image, or a mental representation of sounds, and on the other a concept, or mental representation of meaning. Thus the sign tree, to return to our earlier example,
united an acoustic image, which we might now represent as [trif], with
a concept tree. Each sign was, to repeat, a unit. In another of its seductive images, the Cours had compared the analysis of a language to cutting a sheet of paper. One cannot cut one side of the paper without
cutting the other; similarly, one cannot distinguish either signifiants, like
[trif], or signifis, like tree, unless, at the same time, one distinguishes
both (Saussure, 1972 [1916]: 15). It was for that reason, as we saw, that
the sign could not, as in ordinary usage, be the signifying element
alone.
But to say that a unit has two faces is to say that, between one face and
the other, there is a relation. Thus, in this example, [trif] stands in a
relation of mutual implication to tree. In Hjelmslevs formulation, one
term in the sign relation is, by definition, an expression (udtryk) and
the other is, again by definition, a content (indhold) (Hjelmslev, 1943:
44). That, to repeat, was simply a matter of definition. In using the word
expression Hjelmslev did not imply that signifiers had any other properties; it refers quite simply to a term in a relation whose other term is
referred to by the word content. In using that word in turn he did not
imply that what was signified had any property other than that of entering formally into a relation whose complementary term is referred to by
the word expression. Thus, in particular, expressions were not in principle acoustic images, and contents were not in principle concepts. But the
relation is in other respects as Saussure, in effect, conceived it. We cannot
talk of units of expression, such as, in the same notation, [trif], except
by virtue of the contents that they express; nor of units of content,
such as tree, except by virtue of the expressions whose content they
are. For Hjelmslev too the relation is one of mutual implication. In his
terminology, which often does not translate very easily into ordinary
usage in English, the sign relation or sign function is a solidarity (45),
in which each term or functive depends on the other.

76

The architecture of a language system

This formulation may seem merely to restate, in a way yet more removed
from the realities of speech, what Saussure had been saying in his lectures
thirty years earlier. But by defining the relation as fundamental, we can
understand the nature of such systems much more clearly.
Let us consider still the sign tree. It is formed by a relation between
expression and content, in which one term is specifically what Hjelmslev
calls a sign-expression (tegnudtryk): let us continue to represent this
as [trif]. The other is specifically a sign-content (tegnindhold), which
we will continue to represent by tree. These terms or functives are paired
symmetrically, and in that sense the analogy in the Cours with two sides
of a sheet of paper is still applicable. But each term is a unit that can then
be related to others in its own domain. On the one hand, there is what
Hjelmslev called a plane of expression, and at this level sign-expressions
can be found to form relations, independently of any among the corresponding sign-contents, with other sign-expressions. Thus, in particular, [trif] can be related to other units of expression, like [frif] ( free)
or [trk0] (try), with which it will be found on analysis to share smaller
elements. On the other hand there is a plane of content, and at that level
a sign-content like tree can be found, as we will see in a moment, to
form relations with other sign-contents. Such relations will again be independent of any obtaining between the corresponding sign-expressions.
There is a point at which analogies cease to be illuminating; but, having
cut up Saussures sheet of paper, we can then in effect take either side
of any of the separate pieces, and study it in comparison with the same
side of any other piece.
Both the cutting and the comparison depend crucially on a procedure
that Hjelmslev called commutation. Let us take, first of all, the expression unit realised, in a phonologists terms, by [trif]. This is distinct, as a
whole, from, among others, the one realised by [h0n]. But on what evidence
are units on this level distinct? The answer is that, in the sign relation, the
exchange of one sign-expression for another carries with it an exchange
of corresponding sign-contents. Thus the exchange of [trif] and [h0n] on
the level of expression must entail, on the level of content, the exchange of
tree and hill. In the same relation, an exchange of one sign-content for
another carries with it an exchange of corresponding sign-expressions.
Thus tree is distinct from hill, on the level of content, because, conversely, their exchange entails, on the level of expression, that of [trif] and
[h0n]. In Hjelmslevs terminology both units in the sign relation commute; and they are separate units only if this commutation test is met. As
he presented it, the test applies equally and symmetrically on both planes.
As for the initial cuts, so for all subsequent analysis. Let us compare,
for example, the expression units [trif] and [frif]. In part they are similar

Expression and content

77

([rif]), in part different ([t], [f ]). But replacing [t] with [f ] again
entails, at the level of content, the replacement of tree, as a whole, by
free. Likewise the exchange of [if] and, let us say, [uf] entails the
exchange of tree, as a whole, and true; that of [r] and [l] ([f lif]) that
of free, as a whole, and either flee or flea. By the same test of commutation, we distinguish units of expression, like [trif], as wholes and we
analyse each, on its own level, into smaller units like [t], [r] and [if].
Hjelmslevs formulation of this test was contemporary with the codification of procedures in phonology (3.2); and, as applied to the analysis
of expression units, it was plainly similar. In a phonologists account, two
sounds must realise different phonemes if they distinguish words with
different meanings. In Hjelmslevs account, two smaller units of expression must be distinguished if their commutation entails a commutation
of contents. But just as sign-expressions can be analysed by this criterion into smaller expression units, so in his view could sign-contents,
by a symmetrical procedure, be analysed into smaller content units. In
Hjelmslevs terminology, units such as [t] or [r] were expression-figurae
(udtryksfigurer). Likewise, on the level of content, we could establish
content-figurae (inholdsfigurer).
Till now, Hjelmslev points out, such an analysis has never been made
or even attempted in linguistics (Hjelmslev, 1943 [1953]: 61). The consequences have, he says, been catastrophic, since, without it, the description of content has appeared unmanageable. But the method, as he
sees it, is exactly the same. Let us compare, for example, tree and bush.
In ordinary terms, they are words whose meanings are partly similar:
both woody plant, but one large and the other small. Therefore, in
Hjelmslevs system, just as tree and free have part of their expression in
common, so tree and bush would, correspondingly, have part of their
content in common. Let us call this shared part x. At the same time their
content would also, in part, be different. Just as tree and free are differentiated, in their sign-expressions, by [t] versus [f ], so these must be
differentiated by parts of their sign-contents. Let us call these parts y and
z. In exchanging y and z, we exchange, on that level, the sign-contents xy
(tree) and xz (bush); and at the same time, on the level of expression,
the sign-expressions [trif] and [bvq].
Hjelmslevs own brief illustration (63f.) is of nouns distinguished by
sex. Ram and ewe are, as wholes, distinct sign-contents; so are man
and woman, boy and girl, stallion and mare, and so on. In a straightforward inventory, these are simply different content units; likewise he
and she, sheep, human being (Danish menneske), child or horse.
But, by the commutation test, we can eliminate the first eight by resolving them into combinations. Ram can be analysed as he-sheep; ewe as

78

The architecture of a language system

she-sheep; likewise man and woman as he-human and she-human,

boy and girl as he-child and she-child, stallion and mare as hehorse and she-horse. For Hjelmslev the exchange of he and she, as
parts of sign-contents, was parallel to that of parts of sign-expressions.
In replacing [if] in [trif] with [uf] we effect, on the associated plane,
a change of tree to true. Likewise, in replacing he in he-sheep with
she we effect, on the associated plane, a change of [ram] to [ juf].
The symmetry between the two sides of the paper, or the two planes
that make up the structure of a language system, was now perfect. But
this was not the only way in which the theory took Saussurean ideas to
an extreme. Sign-expressions and sign-contents, expression-figurae and
content-figurae, all were identified, as we have seen, by their relations to
one another. But did they then have any other properties?
It might have been natural to take for granted that they did. For
Saussure, a signifiant was again the acoustic image of some complex of
sounds; a signifi a concept as an element of thought. For Trubetzkoy,
a signifier in the language structure was constituted by distinctive features of sound (3.2); a signifi, one might analogously suppose, would
be defined by features of meaning. In explaining Hjelmslevs system of
relations, I have myself relied on an implicit link between, in particular,
the sounds transcribed [trif] and the abstraction [trif]. But in Hjelmslevs
own account such links were of realisation only. A sign-expression is
characterised just by its relation to a sign-content, a sign-content just by
its relation to a sign-expression. Quite literally, as Hjelmslev said when
these terms were introduced (45), neither of the terms expression or
content meant any more than that.
In developing this idea, Hjelmslev cited (46) a passage in Saussures
Cours in which, independently of language, thought and sound are
both said to be formless. Independently of its expression by words,
thoughts are a shapeless continuum (notre pense nest quune masse
amorphe et indistincte). There are no divisions between ideas until
language comes into play. Nor are sounds any less shapeless. Sound
substance (la substance phonique) is a plastic material (une matire
plastique) that is in turn divided only to supply the signifiers that thought
needs (les signifiants dont la pense a besoin). A language can therefore be represented by a series of divisions made simultaneously in both
domains (Saussure, 1972 [1916]: 155f.). The linguistics of a language
system does not deal with either separately. It operates in the boundary
area where elements of both orders combine (La linguistique travaille
donc sur le terrain limitotrophe o les lements des deux ordres se
combinent, 157). Their combination, we are then told, produces a form,
not a substance (cette combinaison produit une forme, non une substance).

Expression and content

79

This last sentence is found only in part in the written material on which
the Cours is based (see De Mauros note in the edition referred to). But
the formula is repeated at the end of the chapter. From whichever side
a language is approached, we find a complex equilibrium of terms that
condition each other reciprocally (ce mme quilibre complexe de termes
qui se conditionnent rciproquement). In other words, the language
system is a form and not a substance (169).
What Saussure meant is perhaps not so important. We must again
remind ourselves that this is a book which he did not write. But what
Hjelmslev meant was exactly what that formula says. One cannot talk of
phonic substance, even as shapeless, independently of language systems.
Nor can one talk independently of a substance of thought. The sounds
of speech exist only as a manifestation or projection, in a domain outside the language system, of purely formal distinctions among expression
units. The substance of thought is likewise a projection, in a domain
again outside the language system, of purely formal distinctions among
content units.
To understand this, it may help if we look in a different way at our
analogy, in 2.1, of coinage. I can pay someone ten pounds by, for example,
handing over several bits of metal. But I do not always have to pay in
that way. I might instead write a cheque, or use a credit card, or transfer
the money electronically from my account to theirs. In each case I am
paying ten pounds: that is constant. But all else is variable. In one case
a sum of money is realised, we may say, by coins. In another it may be
realised only in an exchange between computers. So what, independently
of these variable realisations, is ten pounds? It is evidently no more than
a numerical value. It is defined, in a specific monetary system, by its
relation to all other possible sums: as ten times one pound, a hundred
times ten pence, and so on. Coins are one of the physical means by
which it can be transferred; and, in general, coins would not be coins if it
were not for the units of value that they represent. But there is no necessary connection. We can easily imagine a system of money with no coins.
As for money so, though the analogy is not Hjelmslevs own, for language. On the expression plane the role of sounds, like that of coins, is
contingent. A distinction between expression-figurae might be realised,
or manifested, by [t] versus [f ]. But the mode of realisation is again a
variable. Writing, in particular, is an alternative substance in which the
same units would be realised by marks on a flat surface: t versus f ,
or T versus F. In Hjelmslevs definition (Appendix, 52), substance is
the variable in a manifestation. Only the system is constant, and it is for
him no more than a network of relations, or Saussurean differences,
among units.

80

The architecture of a language system

By a parallel definition, form is the constant in a manifestation

(Appendix, 51). We might therefore say that, in Saussures Cours, form
was constituted by a relation between constant mental images of sounds
and constant concepts. But just as sign-expressions and expression-figurae
have no intrinsic connection, at their level, with sounds, so, in Hjelmslevs
system, sign-contents and content-figurae have, at their level, no intrinsic
connection with concepts or thoughts. Thought too is a substance, and
as such contingent. Form, in content as in expression, is a system of
relations only.
In summary, then, a language is structured on two planes: plane A, let
us say, which Hjelmslev called that of content; plane B, which he called
that of expression. But plane A and plane B would be perfectly
adequate, since these are merely terms in a bilateral implication. The
form of plane A (content) is a network of relations, which structures a
substance; the form of plane B (expression), likewise. On each level,
substance implies form: there could not be a substance without the form
that structures it. Form does not, conversely, imply substance. It exists as
a pure system of relations: like, for example, those of mathematics. A
substance, in contrast, exists only as the projection of a form, and it is
only when the substances of different languages are compared that we
can talk at all of, for example, speech sounds in general or of thought in
general. In a passage that we have not yet covered, Hjelmslev defined a
purport (Danish mening) that can be abstracted from the content
substances of different languages. Thus there is a purport, in his illustration, common to the thought substances of English I do not know, of its
French translation Je ne sais pas, of the Danish Jeg vd det ikke, and
so on (46f.). There was similarly a purport, though the term is now
acknowledged to be unusual, that can be abstracted from different sound
substances (51). But neither content-purport nor expression-purport
exist independently of language systems.
The symmetry of Hjelmslevs scheme immediately attracted comment.
The main idea, in the words of the Polish scholar Jerzy Kurynowicz, was
to pinpoint structural features common to the two planes (Kurynowicz,
1949: 48). The method Hjelmslev had envisaged was that of an internal
comparison (60), establishing what Kurynowicz describes as a resemblance in form, or a remarkable isomorphism (48), between them. Though
Hjelmslev did not himself speak of isomorphisms in that sense (compare Hjelmslev, 1943: 100f.), the term has often been taken to characterise his and similar theories.
What is perhaps less obvious than the symmetry itself was the role of
abstraction in achieving it. If by expression we mean sound and by
content we mean meaning, there is, on the face of it, no parallel between

Expression and grammar

Phonology
content

81

these domains and no way, in particular, of manipulating meanings as

the commutation test required. The symmetry was only possible in that
Hjelmslev was in fact manipulating neither sounds nor meanings, but
purely abstract counters which were established solely as terms in the
relations that the test defined. Nor, on the face of it, do sounds and
meanings correspond in any simple way. But the solidarity between
contents on our plane A and expressions on our plane B had nothing
literally to do with either. The relation of sign-contents to the purport
of thought was indirect: it had to be if different languages, with different systems at that level, project substances that are correspondingly
different. The relation of sound purport to sign-expressions, whether
similar or not, was also indirect. The connections in general between,
literally, sound and, literally, thought or meaning, were accordingly very
complex. But by abstraction to a system of pure form the mediating
sign relation could be made very simple.
5.2

Phonology and grammar

Glossematics was debated actively in the years that followed the Second
World War, especially after Hjelmslevs account was published in English. But by the time of his death in the 1960s its influence was no longer
direct. The treatments that were dominant then had different origins, and
had been developed with the full insight of synchronic phonology, either
of the Prague School or of the Bloomfieldians, behind them. They are
also more complex.
Let us begin by looking back to a scheme inherited from antiquity. In
a formula that appears to have originated with the Stoic philosophers,
language is, in general, articulated sound. The term articulated is from
the Latin word for a joint (articulus); and what was meant is that, in
distinction as it was then thought to the cries of animals, or to other
inarticulate sounds emitted by human beings, speech was organised into
distinct units. Its smallest unit was the letter (Latin litera): as we have
already noted (3.1), this was traditionally a spoken and not just a written
unit. Thus, in a standard formula transmitted in the fourth century AD
by the grammarian Donatus, the letter is the smallest part of articulated
vocal sound (pars minima vocis articulatae) (in Holtz, 1981: 603). Other
units form a hierarchy in which units at each level are composed of units
at the level below it. Letters first combine to form a syllable (Latin syllaba). For example, in English the letters s and a form the syllable
sa. Syllables then combine to form a word (Latin dictio). For example,
sa and tin form the word satin. Finally, at the highest level, words
combine to form an utterance or sentence (Latin oratio). Of these units,

82

The architecture of a language system

a word or a sentence also has a meaning; a letter or a syllable does not.

But that apart, they are related throughout by simple composition. As
Priscian put it a couple of centuries later, just as letters combining
appropriately form syllables, and syllables words, so too do words form
a sentence (quemadmodum literae apte coeuntes faciunt syllabas et
syllabae dictiones, sic et dictiones orationem) (Priscian in Keil, 18559,
II: 108).
For the ancient letter we can read, by the 1930s, the phonologists
phoneme. But, in addition, it had long been clear that words can be
divided into other units. Take, for example, unkindnesses. We might imagine, once more, that we are investigators to whom the structure of
English is unknown. So, in a form such as his many unkindnesses we will
want to work out where, as Bloomfield put it (2.2), there are recurring
likenesses in form and meaning. We might find that many recurs in, for
example, her many friends or They were many; also that, holding his
unkindnesses constant, we can replace many with few (his few unkindnesses),
or indeed remove it (his unkindnesses). On such evidence, we establish
that his, many and unkindnesses, as wholes, all meet our criteria. But we
will find that they are also met by smaller units. The form kind recurs in
He was kind or a kind man; un- can also be removed (his many kindnesses);
as we removed many so we can also remove -es (his unkindness); -ness
recurs in, say, his sadness or his happiness. On such evidence un-, kind,
-ness and -es can be identified as other recurring likenesses.
Since the Renaissance, units like un-, -ness or -es had been described
as affixes, and in a more recent tradition, which was to persist among
French linguists in particular, were one kind of (to use for the moment
the French term) morphme. Thus, for Joseph Vendryes, morphmes
were elements that expressed relations among ideas. In the same example, kind is a root or (again in French) a smantme. Smantmes
were said to represent ideas themselves: thus, in this case, it would be an
idea which all words that include it, such as kind or kindness or unkindness, have in common (compare Vendryes, 1968 [1923]: 92f.).
For Bloomfield, and most other linguists since the 1920s, un-, kind,
-ness and -es are indifferently morphemes. But, whatever the terminology, this result must clearly complicate the ancient hierarchy. Take, for
example, uneasy. It has three successive syllables: in the usual analysis,
[k], [nif] and [zi]. It can also be analysed into three successive morphemes:
un- ( [kn] ), eas(e) ( [ifz] ) and -y ( [i] ). But at no point in this word do the
boundaries of syllables and morphemes coincide. We can say once more
that its component phonemes are combined in syllables and the word in
turn is formed from these. We can say instead that they are combined
in roots and affixes, and the word is then composed of those. But in this

Expression and grammar

Phonology
content

83

word as in many others, syllables and morphemes cannot both be units

in a single hierarchy.
The obvious answer is that, within words at least, there are two hierarchies. As the ancient grammarians had said, speech is articulated into
units like [k], [n] or [if]. The phonemes that these realise form syllables,
which like them are units without meaning. But speech also realises, and
is in that sense articulated into, units like un-, eas(e) and -y, which,
independently of syllables, form words. As letters or phonemes are the
smallest units that do not have meaning, morphemes, at the lowest level
in a parallel hierarchy, are the smallest that do.
In Europe at least, the classic account is that of Andr Martinet. Suppose, to follow his argument from the beginning, that one has a headache.
One might perhaps groan, which, as ancient writers would have put it, is
to emit a sound that is inarticulate. But one might also, if one is French,
say jai mal la tte I have got a headache. Unlike the groan, this
is articulated into distinct units, written j(e), ai, mal and so on, each
identifiable, like the un-, kind or -ness in English, by their recurrence in
other utterances. In the tradition of Saussure and the Prague School,
each is a minimal linguistic sign. Within each, a specific vocal form
(thus, for the sign mal, the form [mal] ) is linked to a specific meaning (in
Martinets notation, mal). I am citing from a textbook published at the
end of the 1950s (Martinet, 1970 [1960]: 13f.), in which minimal signs
were, in his terminology, monemes (monmes) (15).
The division into minimal signs was Martinets first articulation;
and its practical advantage, as he pointed out, is that by combining discrete units of meaning, which together number no more than some
thousands, speakers of a language can communicate many more things
than they could by millions of unarticulated groans, cries, and so on. But
then, at a subsidiary level, vocal forms like [mal] are themselves not
shorter groans or cries; but are articulated, in turn, into phonemes. Thus
the vocal form [tet], which is the signifying side of the sign tte, is articulated in turn into the phonemes [t], [e] and again [t]. This was Martinets
second articulation; and its practical advantage, once more, is that, by
virtue of it, a small number of these ultimate sound units can combine
to distinguish the thousands of vocal forms that signify, or have meaning, in the first (15). Thus the initial [t], on its own, distinguishes [tet]
from, for example, [bet] (bte); the final [t], on its own, distinguishes it
from, for example, [tel] (tel), and so on. If language were not articulated
on both levels, it would not be nearly so efficient.
These important insights were first set out clearly at the end of the
1940s (Martinet, 1949). But they were also central to the work of
Charles F. Hockett and the other Bloomfieldians. In his own much longer

84

The architecture of a language system

introduction to linguistics, Hockett pointed to seven properties that

potentially distinguish human language from communication in other
species (Hockett, 1958: 574ff.). The first was a similar property of duality or duality of patterning. It would be inconvenient, as he pointed
out, if our system of communication were not structured in this way;
and, of his seven, it was the only property that he could discover, on the
evidence then available, in man only.
In any language, as Hockett conceived it, an utterance consists of
an arrangement of the phonemes of that language; at the same time, it
also consists of an arrangement of the morphemes of that language.
Any sentence would accordingly be structured on two levels. On one
level morphemes, which were the nearest Bloomfieldian equivalent to
Martinets monmes, would be related syntagmatically (in Saussures
terminology) to other morphemes. Such relations were the nearest equivalent of what Bloomfield called grammatical forms (Bloomfield, 1935
[1933]: 264); and, for Hockett, morphemes and their arrangements constituted the grammatical system. On another level phonemes would be
related syntagmatically to other phonemes. This was the equivalent of
Martinets second articulation; and, for Hockett, phonemes and their
arrangements constituted the phonological system. We can see that,
despite differences in terminology, both scholars were, in these respects
at least, describing components or levels of a language structure in
essentially the same way.
In linguistics, Hockett had already declared in his mid twenties, there
are . . . two basic levels, phonological and grammatical (Hockett, 1942:
3). One has the phoneme as its elementary unit; the other the morpheme.
But what exactly was the nature of this second unit in -eme, and how
was it related to the first?
We have seen Martinets answer. A morpheme, or rather a moneme,
was a sign in Saussures sense. One side was a signifier, and was analysable directly into phonemes. Martinets account agreed thus far with
Hjelmslevs, and with others generally in continental Europe. It was also
not unlike, in spirit, Bloomfields. But in Bloomfields system of terminology, a morpheme had been, in Saussurean terms, a signifier only. It was
a unit that had a meaning, not one that itself consisted of both a form
and a meaning. For Hockett and the other Americans of his generation it
was a unit of form, in Bloomfields sense, only; and, in one extreme view,
meaning was to play no part in its definition. Nor could we assume that,
when a morpheme had been identified, any single meaning could in reality
be associated with it. Nor, finally, was a morpheme composed of phonemes. We have seen that, by the end of the 1930s, phonemes were not
literally speech-sounds. They were instead abstractions realised by sounds

Expression and grammar

Phonology
content

85

(3.2). By the end of the 1940s, morphemes were a still more abstract unit,
not composed of phonemes but in turn realised through them.
This view developed very quickly, and to see why we must think back
to the period in which it was formed. The theory of the phoneme was still
new and still very exciting; and in America especially a method had been
developed by which the phonemes of any language could be rigorously
identified. The criteria were objective: different linguists could, in principle, apply them independently to the same material and obtain equivalent results. The next step was to develop rigorous criteria for other units.
Just as, at one level, an utterance could be analysed, exhaustively so far
as a linguist was concerned, into phonemes, so at another level it could
be analysed exhaustively into, in Bloomfields sense, morphemes. How
could these in turn be objectively identified?
First we needed criteria for segmentation. Why, for example, is
unkindnesses to be divided into un-, kind, -ness and -es, and not, say, into
unk-, -indne- and -sses? Then, at some level, we had to identify invariants.
In words like unkindnesses or peaches ( [piftq0z] ) the plural ending, to
return for the moment to older terminology, is [0z]. In, for example,
apples or pears it is [z] ( [aplz], [ptfz] ); in nuts or currants it is [s] ( [nkts],
[kkrints] ); in irregular plurals, such as children or men, we find yet more
variation. But behind it, all grammarians will agree, there is the constant
or invariant that we are calling plural. How were we to identify that,
and what sort of unit is it? In Come in, the verb is phonetically [kkm]; in
He came in, the past tense of the same verb (to come) is phonetically
[ke0m]. But all grammarians will agree that these are indeed forms of the
same verb. How was that unit (to come) to be identified? And, again
what sort of unit? What was it?
In Martinets account, the signifi was constant but was realised in
forms that vary (se manifest[e] . . . sous des formes variables) (Martinet,
1970 [1960]: 102). The variation is thus in the signifiant. Martinet also
talked of variants of the signifying side of monemes (variantes des
signifiants de monmes) or, more briefly, variants of monemes (variantes
de monmes) (106). By implication, the moneme itself was constant.
Thus, to take his own example from French (102), aller (to go) is a
signifi realised variously in the forms [al] (aller), [va] (va), and so on. By
implication, [al] and aller form one variant of a moneme, [va] and
aller a second variant of the same moneme.
For the Americans, whose work Martinet knew well, the invariant had
to be the morpheme. In phonology, as we have seen, the phoneme was a
class of allophones (end of 3.2) in complementary distribution. But our
variant endings, for example, are also in complementary distribution.
The plural [0z] is found only after a certain set of consonants ( [tq ] in

86

The architecture of a language system

peaches, [s] in unkindnesses, and so on); and, after these consonants, we

do not find either [s] or [z]. Where [s] is found, after among others the [t]
of nuts or currants, we do not find [z]; nor vice versa. The irregularities in
men or children are precisely irregularities, found only in specific words.
The parallel was, by the later 1940s, obvious. In a phonologists terms,
the procedure by which we establish phonemes involved, first, the segmentation of recorded utterances into phones; then the classification
of phones as allophones of their respective phonemes. Each utterance
could then be represented as a sequence or, more generally, arrangement
of phonemes. The next step after that was to segment such representations into smaller arrangements such as, in the case of peaches, [piftq]
and [0z]. In the terminology that Hockett now proposed, these were analogously morphs (Hockett, 1947). In a further step morphs would be
classified, by among others a criterion of complementary distribution,
as allomorphs of their respective morphemes.
A further criterion was, on the face of it, that allomorphs should
share a meaning. Thus there is a meaning (Martinets signifi) common
to come and came, or to French va and aller, or to all the plural endings.
But the morpheme itself was now identified by a procedure operating
solely on arrangements of phonemes. If defined by that procedure, it was
again a unit that had meaning, not a unit, like that of Martinet and
others in the Saussurean tradition, that was itself, in part, of meaning.
But neither was it a unit composed of phonemes. Come, we could now
say, represents one morpheme, realised in both [kkm] (come) and [ke0m]
(came). The category plural was another morpheme, realised by [0z]
and so on. But only their realisations could be divided into parts; the
morphemes themselves are formally primitive.
These were technical details, and a reader may well feel that, in a
history that has hitherto dealt largely with more general ideas, they might
helpfully have been skipped over. But the result is fundamental to most
later theories of a language system. In earlier treatments, units like
Saussures signifiants, or morphemes as Bloomfield had represented
them, had been related directly to signifis or units of meaning. Even in
Martinets, a variant like [al] in aller was still directly related to aller.
But the morpheme as defined by Hockett had the properties of neither an
expression nor a content. It was not Saussures mental representation
of sounds; nor made up of distinctive features of sounds; nor even of
Hjelmslevs expression-figurae. It could indeed be realised by allomorphs,
like [va] and [al] in French, which were phonetically very different. At the
same time it was not a unit of meaning: not even Hjelmslevs abstract
combination of content-figurae. It was simply an unanalysed whole, by
which sounds and meanings were related through a mediating level. As

Expression and grammar

Phonology
content

87

a unit on that level, come can be related, on the one hand, to a form
[kkm]; on the other, to a meaning represented for the moment by come.
But these relations are separate, and there is no longer a direct link
between [kkm] and come.
The entire scheme was set out in Hocketts general introduction to
linguistics, ten years after the article I have referred to. A language was
as a whole a system, which can be broken down, he said, into five
principal subsystems (Hockett, 1958: 137). Three were central: a grammatical system and a phonological system, and a third which linked
them. Two others were peripheral. When Hockett was writing, the term
interface was not yet in fashion; but the peripheral subsystems formed
two opposite interfaces with domains outside language.
Morphemes, such as plural or come, were the basic units of the
grammatical system. It consisted, therefore, of a stock of morphemes,
and the arrangements in which they occur. At this level, then, the word
peaches may be represented by the arrangement peach + plural; came by
the arrangement of come with, as a further morpheme, past tense. The
phonological system was, in parallel, a stock of phonemes, and the
arrangements in which they occur. So, at this level, peaches and came may
again be represented by arrangements of [ p], [if], and so on. It was by
virtue of these systems, then, that language has the property, as we saw
earlier, of duality. The third central component was the code that ties
together the grammatical and the phonological system. By this system,
peach as a single morpheme would be related to the arrangement of
phonemes [piftq] and, by reference to the [tq], plural, as a morpheme
following it, would be related by the code to [0z]; an arrangement of
come with past tense would in the same sense be related as a whole to
[ke0m]. Since this code related morphemes to phonemes Hockett called
it the morphophonemic system.
Finally, one interface was formed by a peripheral semantic system,
which associated the units and arrangements of the grammatical system
with things and situations, or kinds of things and situations (138). Meanings, therefore, were associative ties (139) between linguistic units and
the world that speakers are talking about. The opposite interface was
formed by a phonetic system which related arrangements of phonemes
to produced or perceived sounds.
For a final twist to this part of our story we can return to the ancient
scheme with which this section began. For the ancient letter we can
again read phoneme; and by the 1930s, as we have seen, phonemes were
already abstractions. At that stage, however, it could still make sense
to say that words and sentences were sequences of phonemes. In the
schemes developed by the end of the 1950s this was drastically modified.

88

The architecture of a language system

The phoneme, as the basic unit of one level of articulation, was matched
by the morpheme, or by Martinets moneme, as the basic unit of another.
At this new level of abstraction, sentences were formed ultimately of
morphemes or of signs corresponding to them. Thus, following Hockett,
a sentence The women came in might be represented at that level by an
arrangement of morphemes the + plural + woman . . . and so on. But
what then of, in particular, the word? Traditionally the sentence would
have, as its smallest units of meaning, four words. But if morphemes
or monemes now had that role, did this traditional unit still have a place
in the system?
Martinet was one structuralist who thought not. In the introduction I
have been citing he put the term word (mot) in inverted commas, and
argued that it should be discarded (Martinet, 1970 [1960]: 114ff.). Take,
for example, French nous courons we are running. By tradition, this is
a sequence of two words: nous we and courons, which is a first person
plural of the verb courir to run. The first person plural is distinguished
by the ending -ons and, still in the traditional account, the verb agrees
with nous as its subject. But for Martinet (104) the monemes that make
up this form were just two. One was, as we might expect, cour-: that is to
say, the signifiant [kur] in conjunction with the signifi courir. The
other united a signifi first plural neither specifically with [nu] (nous)
nor specifically with [] (-ons), but with a discontinuous signifiant,
[nu . . . ], that includes both. In that way the analysis cut clean across
the traditional boundary between words, which was therefore invalid.
In the next section (105), Martinet also put agreement (accord) in
inverted commas.
Not everyone was an iconoclast to that degree. But the logic of such
treatments could not be ignored. If the phoneme and the morpheme were
the first great technical gains that structural linguists might claim, the
word was beginning to look like the first casualty.
5.3

Deep structure and surface structure

The British linguist C. E. Bazell remarked in the 1950s that, if it is

possible to discover any aim common to all linguistic schools, it was the
reduction, by terminological devices, of what he called the fundamental
asymmetry of linguistic systems. If, for example, there are phonemes and
allophones it is assumed that there must also be morphemes and
allomorphs. If there is a form and a substance of the expression, then
there must also be a form and a substance of the content. If phonemes
can be analysed into relevant phonic features, then morphemes (as we
will see more clearly in a later chapter) were analysed into relevant

Expression
Deep
structure
and and
content
surface structure

89

semantic features. No single scholar might subscribe to all specific

proposals current, but the relation between them was unmistakable
(Bazell, 1953: iii).
These were the comments of a critic, and most structuralists would
have protested that the symmetries which they saw between the two sides
of the linguistic sign, or the parallels that they explored between descriptive method in phonology and in grammar, were not figments. But the
passion for isomorphism, as Kurynowicz had described it, was then rampant; and, if Bazell had been writing towards the end of the next decade,
he could have added yet more instances to his list. Of the new schemes in
the 1960s, the most influential by far was that of Noam Chomsky. When
he proposed it he was still in his late thirties, and much of both its
substance and its terminology has since changed. But Chomskys reputation, in linguistics and still more outside it, had then reached its zenith.
This scheme was therefore widely known and widely pondered, often by
commentators who knew little or nothing of those that had preceded it.
Its sources lay, in part, in Chomskys earlier account of syntax. By
syntax grammarians have traditionally meant the study of relations
among units that make up the sentence. Originally these were words:
thus, to describe the syntax of The women came in, a grammarian would
relate the noun women to the article the, the verb came to the preposition
or adverb in, and so on. But by the 1950s they were widely taken to be
morphemes. In the same example, the and in were indivisible units; but
women and came could both be represented by two: woman + plural,
come + past. Syntax was accordingly concerned with the arrangements,
in Hocketts sense, of morphemes. Hocketts scheme was published later
than the earliest work by Chomsky. But Chomsky had been a protg
of Zellig Harris, and these results were taken for granted by most structuralists of the American school.
How then were morphemes arranged in sentences? The answer given
was, in part, that they form hierarchies of intermediate units. Thus, at
the lowest level, woman and plural form one intermediate unit which
we write as women. We can show this, in a common notation, by enclosing them in square brackets: [woman plural]. Both larger units and the
morphemes that compose them are of various classes. Thus women is, in
the traditional classification into parts of speech, a noun. We may show
that as follows: N[woman plural]. In this notation, which translates easily
into others, successively larger units are enclosed throughout in brackets
and their class is indicated by a symbol, such as N for noun, subscript
to them.
In the same way come and past tense form a verb: V[come past]. But
then, in our example, that in turn will form a unit with in: [ [come past] in].

90

The architecture of a language system

Traditionally this is a predicate, and predicates were said to have verbs

as essential elements; therefore they were described in the 1950s as verb
phrases. So, with a subscript VP, VP[ V[come past] in]. In the same way
units like the women were described as noun phrases: NP[the N[woman
plural] ]. Finally, in a still larger pattern of arrangement, the sentence
as a whole (S) would then be made up of the noun phrase plus the
verb phrase: in summary, S[ NP[the women] VP[came in] ]. In this way, any
sentence could be represented by a sequence of morphemes, such as
the + woman + plural . . ., combined, stage by stage, into a hierarchy of
larger units.
The basic insight precedes Chomsky. But it was in his work in the
1950s, and in particular in his monograph Syntactic Structures, that the
nature of this system was most rigorously explored. To describe the syntax of a language was to set out, in a series of statements as simple and as
general as possible, all the ways in which a unit of one class can combine
with units of others. Thus by one statement, or in Chomskys terms one
rule, a sentence (S) can, in English, consist of a noun phrase (NP)
followed by a verb phrase (VP). By another rule, a noun phrase can
consist of an article, such as the, followed by a noun; by another, a noun
may consist of a morpheme like woman followed by the plural morpheme; and so on. I referred briefly, at the end of 2.2, to the concept of
a generative grammar. A grammar taking this form could be seen precisely, in the way that Harris had foreshadowed, as a set of instructions
by which sentences are generated (again Z. S. Harris, 1954: 260). If we
begin with S as the largest unit, we could take the first of these rules as
an instruction to derive a more detailed structure NP + VP. We could
take the second as an instruction to derive from that a structure in which
NP is in turn divided: article + N + VP. By successive steps we would
eventually reach a point at which all elements are classes of morphemes.
Any sequence of morphemes which would satisfy that structure represents one utterance that is possible in the language.
In Chomskys account, a generative grammar in which all rules were of
this form was, more precisely, a phrase structure grammar (Chomsky,
1957: 26ff.). But, having made their form explicit, he then argued that, on
their own, they were inadequate. He did not know (34) whether English,
for example, literally could not be generated in this way. But, as soon as
we consider . . . sentences beyond the simplest type and, in particular, . . .
attempt to define some order among the rules that produce them, we
run into numerous difficulties and complications (35). The remedy, as
he saw it, lay in adding rules of a quite different kind, by which one
hierarchical structure, like that of The women came in, is transformed
into another.

Expression
Deep
structure
and and
content
surface structure

91

Let us take a classic illustration, which concerns the activepassive

relation (42). In, for example, John eats lunch the verb eats, which is
traditionally active, had two elements: the eat morpheme plus, in
Chomskys notation, an S morpheme. But in Lunch is eaten by John the
form is eaten, which is traditionally passive, had two more: first be,
which now combines with S to form is; then en, combined with eat
and realised, in this instance, by -en. These further morphemes went
together (39). They could not be added when a noun or noun phrase, like
lunch, follows directly: thus one cannot say John is eaten lunch. Neither
could they be omitted, or replaced by others, in a context such as Lunch
by John. One cannot say Lunch eats by John; nor, for example, Lunch
is eating by John. Finally, in a full-fledged grammar, we would need
to place restrictions on nouns in the positions of lunch and John. Such
restrictions would permit, for example, an active sentence such as John
admires sincerity or Sincerity frightens John, while excluding the inverse non-sentences Sincerity admires John or John frightens sincerity.
Likewise, in the passive, they would have to permit both Sincerity is
admired by John and John is frightened by sincerity, while excluding
both John is admired by sincerity and Sincerity is frightened by John.
If we try to include passives directly, alongside actives, in a phrase
structure grammar, we will have to state all such restrictions once, for
one construction, and then restate them all, in the opposite order, for
the other (43).
This inelegant duplication, with all the special restrictions affecting
be and en, can be avoided only, Chomsky argued, if we introduce a
new rule by which, for any active sentence, there is automatically a corresponding passive sentence. So, given John eats lunch, with a hierarchical
structure of the sort already outlined, it will follow that there is another
sentence Lunch is eaten by John, whose structure will in part reflect it. By
this new rule be and en would be obligatory: therefore there cannot
be a passive construction, like Lunch eats by John, that does not include
them. They would not be specified, as a unit, by the rules for actives:
therefore there is no active sentence John is eaten lunch. If there is an
active John admires sincerity, it followed that there is a passive Sincerity
is admired by John. Since passives were formed only by this rule, then,
by excluding an active Sincerity admires John, we also exclude John is
admired by sincerity.
This new rule was a grammatical transformation (44), and a phrase
structure grammar plus transformations was a transformational grammar. Its attraction, once more, was that in this way we could avoid
inelegancies, complications, duplications, and so on. But Chomsky also
claimed that such rules would explain some aspects of meaning. This was

92

The architecture of a language system

argued, in particular, for examples of what was often called grammatical

ambiguity.
A famous example was the shooting of the hunters (Chomsky, 1957:
88f.). This can, in Chomskys words, be understood ambiguously, in
a sense which is either like that of the growling of lions (the hunters are
doing the shooting, the lions are doing the growling) or like that of the
raising of flowers (the hunters are being shot, the flowers are being raised).
For other grammarians these were examples of subjective and objective
genitives: the hunters is related to shooting in the sense of either a subject (compare The hunters shoot) or an object (compare They shoot the
hunters). For Chomsky, therefore, they reflected different transformations
by which forms in -ing could be derived. Careful analysis . . . shows that
we can simplify the grammar (89) if we set up rules by which, in one case,
any sentence like The lions growl would carry . . . into the corresponding phrase the growling of lions; and, in the other, any sentence like They
raise flowers would carry into a superficially similar the raising of flowers.
In what Chomsky had earlier called a full-fledged grammar, there would,
by implication, be restrictions which exclude non-sentences such as The
flowers raise or They growl the lions. But both The hunters shoot and
They shoot the hunters would carry, by different transformations, into
the shooting of the hunters. This was, in Chomskys words, a clear and
automatic explanation (88f.) of the ambiguity.
That is as far as Chomsky had got by 1957. The parts of a generative
grammar were therefore, first, a set of rules for phrase structure; then a
set of transformations; then, finally, a morphophonemic part which,
like Hocketts morphophonemic system (5.2), related abstract representations of words, such as be + S or admire + en, to sequences
of phonemes such as 0z or idmk0id (Chomsky, 1957: 45f.). In terms
of some more general theory of language (102) we could also study
correspondences between the formal structure of sentences, as generated
by such a grammar, and features of meaning.
But in the 1960s it was assumed that grammars should themselves give
an account of meanings. It is quite obvious, Chomsky said in one work
of this period, that sentences have an intrinsic meaning determined
by linguistic rule (Chomsky, 1973 [1967]: 115). To know a language was
to control a system constituted by rules that interact to determine the
form and intrinsic meaning of its sentences (1973 [1968]: 71). A generative grammar was a system of rules that relate [phonetic] signals to what
are called semantic interpretations of these signals (1966: 12). Such a
grammar would again have parts or components (Trubetzkoys Teilsysteme). So what exactly were the levels at which sentences would be
represented?

Expression
Deep
structure
and and
content
surface structure

93

Two were seen as given. If a grammar relates signals to meanings of

signals it must, first, represent the signals themselves. These were not
Chomskys main concern; but a natural proposal, developed by his colleague Morris Halle, was that successive segments should be represented,
in the tradition of the Prague School, by phonetic features. Halle had
been a colleague of Roman Jakobson, who had escaped from Europe to
America in the Second World War; and in the 1950s had contributed to
a theory of phonology in which phonemes were defined as bundles of
concurrent features (Jakobson and Halle, 1956: 5). In his representations
of signals segments correspond more nearly to phones (3.2); except that,
as Chomsky and Halle make clear in a joint work, grammars were concerned not with physical sounds, but with signals as they are mentally
constructed by the speakers of a language (Chomsky and Halle, 1968:
14). The features of which they were said to be composed refer by 1968
to postures of the vocal organs by which speech is produced. But that
aside, such representations were in the tradition of Saussures acoustic
images (2.1).
As a grammar must by definition represent the signals, so, again
by definition, it must represent semantic interpretations. The problems
here were still veiled, as Chomsky put it, in their traditional obscurity
(Chomsky, 1966: 13). But in the same passage he referred to work by
colleagues, notably Jerrold J. Katz, which implied that meanings might
be intrinsic if they did not reflect the particular circumstances in which
sentences are uttered. Semantic theory was, in the words of Katz and
Jerry A. Fodor, a theory of the semantic interpretation of sentences in
isolation (Katz and Fodor, 1963: 177). We will return to their theory in
another chapter. But when a sentence is uttered it will be in a specific
setting (176ff.), and will have its meaning in that setting. Its meaning
in isolation was one assumed to hold independently of settings.
A generative grammar would then mediate between these, just as, in
Hocketts design of a language (1958: 137), central systems mediate
between peripheries. How then did the parts that Chomsky had established in 1957 fit in?
The answer was superbly simple. As Chomsky expounded it, the next
step was to posit a neutral technical notion of syntactic description
(Chomsky, 1966: 13). This would mediate the relation between forms and
meanings, in that any such description would uniquely determine both
a semantic interpretation and a phonetic form that corresponds to it.
So, whatever the intrinsic meaning of, let us say, She was disturbed by the
growling of lions, it would be determined by a representation of the
syntax of this sentence which would also determine that, phonetically,
its first segment is [q], its second [if], and so on. A sentence like, say, She

94

The architecture of a language system

was disturbed by the shooting of the hunters would, naturally, have two
semantic interpretations. These would accordingly be determined by
two different syntactic descriptions, each, however, determining the same
phonetic signal.
The form of a syntactic description is now sufficiently obvious. In the
first of our examples the structure of a sentence like The lions growl
carries through, as Chomsky had argued in the 1950s, into that of the
growling of lions. The structure of an active, like The growling of lions
disturbed her, was in addition transformed to that of a passive. Its syntactic description would accordingly relate two representations. One was
an initial structure generated only, following Syntactic Structures, by rules
for phrase structure. This would include a smaller structure in which lions
is the subject of growl: [lions growl ]; and, although this had the form of a
sentence (S[lions growl ] ), it would occupy the same place as a noun phrase
in, for example, The noise disturbed her. It was in that sense both a
sentence and a noun phrase: NP[ S[lions growl] ]. The initial structure as
a whole would thus be one in which this smaller sentence is in turn the
subject of disturbed.
The other representation was the structure that, by transformations,
this would carry into: thus, in part, S[she was disturbed by NP[the growling
of lions] ]. But two further points will now be clear. First, it was this
second representation that specifically determines the phonetic form of
the sentence. To be exact, a grammar would include rules that relate she
to a phonetic representation [qif], be + past to [wiz], and so on, in the
order in which successive morphemes appeared in it.
Likewise, it was the initial representation that, to all appearances,
determines the semantic interpretation. In our second example, the
shooting of the hunters would be related to initial structures in which
hunters was variously a subject (NP[ S[the hunters shoot] ] ) or an object
(NP[ S[someone shoot(s) the hunters] ] ); and, on that basis, we could posit
rules by which it was interpreted in different ways. But the structure
that was derived by transformations would be in either case the same.
Therefore, while the latter would determine the phonetic signal, it would
seem that it was irrelevant to the meaning.
In Chomskys famous terminology, the semantic interpretations were
determined by deep structures. As he defined it at this stage in the
development of his theories, the deep structure of a sentence was precisely the aspect of a syntactic description to which, by the rules of what
he called the semantic component of a grammar, a representation of
meaning was assigned (Chomsky, 1966: 16). Phonetic representations
were instead determined by surface structures. The surface structure of
a sentence was thus correspondingly the aspect of a syntactic description

Expression
Deep
structure
and and
content
surface structure

95

to which the rules of a phonological component assigned what were

also called phonetic interpretations. This phonological component was
in part the equivalent, it can be seen, of his or Hocketts earlier morphophonemics. The central part was a syntactic component that, as in
Syntactic Structures, included transformations. This generates [syntactic
descriptions] each of which consists of a deep structure and a surface
structure.
The work I am citing is the published version of a series of lectures
delivered in 1964. But the scheme was summarised in Chomskys Aspects
of the Theory of Syntax (Chomsky, 1965: 16), and it was from that
account, in a chapter that in later sections dealt inspiringly with issues
in the philosophy of language and the general history of ideas, that it first
caught everyones imagination. It was therefore tempting to read into
the term deep, in particular, far more than was actually there. But its
interest is again as part of a technical scheme of levels, by which, as in
Hocketts, a relation between forms and meanings could be mediated.
Famous as it was, the scheme was of the 1960s and within ten years was
already drastically revised.
But, as a scheme of the 1960s, it is of great historical importance. With
many others, it assumed that meanings can be abstracted as invariants:
the basic relation was accordingly between an abstract level of phonetic
representation (Chomskys signals) and a similarly abstract level of
semantic interpretation. These levels interfaced, by implication, with
real sounds and real meanings. But semantic representations were not
just the signifis of individual signs. They were the meanings, as determined by a generative grammar, of whole sentences. Nor were parts of
meanings related one to one to parts of signals. Phonetic representations
were paired first with surface structures; and this relation was already not
as simple as sign theories had at first implied. Surface structures were in
turn paired with deep structures; and the arrangements of morphemes
on these levels, to borrow again Hocketts terminology, could be as different, both in the units themselves and in their order, as need be. Finally
deep structures were paired with semantic representations; and, when
the veils of obscurity were lifted, that relation might be as complex. For
any sentence, the relation between the signal and its meaning, via the
mediating level of syntax, could in principle be very indirect.
Such schemes were therefore apt to be rejected by any structuralist for
whom the Saussurean link between the two sides of the sign remained
central. By the same token, of course, anyone who took that view would,
from a Chomskyan viewpoint, be concerned with syntax only at a surface
level.

96

Internalised language

Internalised language

By 1960 structural linguists could look back on thirty or more years of

progress. The theory of the phoneme was now widely accepted; and,
although there were disagreements in detail between most Americans
and most Europeans, few disputed that it was one central unit of language. For many linguists the morpheme or a unit like it was another,
and the identification of these units in -eme, as the elements of articulation on two different levels, was increasingly a basis for the practical
description of languages. Such findings had begun to work their way
from technical publications into textbooks. I have referred to Hocketts
general introduction to linguistics (Hockett, 1958). But it had been
preceded by at least one specialist textbook in descriptive linguistics, by
H. A. Gleason, whose table of contents gives a clear view of how students,
in at least part of the English-speaking world, were taught (Gleason,
1961 [1955] ). Martinets introduction, to which I have also referred, was
to have a similarly wide success in continental Europe (Martinet, 1970
[1960] ).
These achievements were ostensibly founded in the thought of Saussure,
in lectures that had last been given in 1911, and of Bloomfield, formulated in its mature phase in the 1920s. A linguist was therefore describing
either an underlying system whose reality was supra-individual (2.1),
or a set of potential utterances (2.2). But both accounts had naturally
reflected the preoccupations of their own day and, with hindsight, though
the structure of a language system had been worked out in some detail,
it was time to ask again exactly what reality such systems had. A new
answer was given by Noam Chomsky, and the Chomskyan revolution
in linguistics, as it has been widely described since the 1960s, rested on
it. How then did this revolution unfold?
In an account in the 1980s Chomsky himself distinguishes two major
conceptual shifts (Chomsky, 1986: 6). The first inaugurated, in his
words, the contemporary study of generative grammar; and consisted
in a shift of focus, as he puts it later in the same book, from a view of
language as behavior and its products to the study of systems of mental
96

Generative grammars

97

representation and computation (51) underlying behaviour. Such systems

of mental representation constitute an individual speakers knowledge
of a particular language, and a generative grammar, as he then defined
it, is a theory of the state in which the mind/brain of a person knowing
a language must be (3). The second conceptual shift is said to be theoryinternal and to be in process as he was writing (6). It led to a specific
theory of genetic factors in the acquisition of this knowledge, or the
development of these systems of mental representation, which had by
then, in the 1980s, become the centre of this thought. Now the theory to
which this was internal is that of, in general, generative grammar. Therefore it is with that notion that we should begin.
6.1

Generative grammars

The term generative grammar had in origin a perfectly transparent sense,

and did not at first bear the construction that, in passages like the one
I have just cited, Chomsky would eventually put on it. We have seen, to
begin, how Bloomfield had defined a language as a totality of potential utterances (again 2.2). Each utterance in a language could then be
represented by an arrangement of smaller units, up to the limits of, again
in Bloomfields account, a sentence (2.3). So, for example, English would
be a set of which the arrangement of words I + (a)m + hungry is one
member. By implication, other arrangements would not belong to the
language. For example, there would be no potential utterance Am hungry
I; or, splitting hungry into morphemes, Im -y hung(e)r. A grammar was
a description of a language; and, under this definition, it described such
a set. It could therefore be interpreted, in the words of Zellig Harris (2.2),
as a set of instructions by which a language is generated.
A set is in a mathematicians definition any collection of members. We
can accordingly define a language in that spirit, in a sense whose application to the real world is potentially much wider than to languages in
the ordinary sense, in which we are interested. The members of such a
language are by definition its sentences, and their number may in principle be either finite or infinite. But each individual sentence is in turn
a finite sequence of elements, and the number of such elements, for any
language, will itself be finite. Now a language as a mathematical object
bears as much relation to a language as conceived by Bloomfield as, for
example, a mathematicians concept of a triangle bears to an actual
object whose shape is triangular. For that reason, I have scrupulously
put both language and sentence in inverted commas. But all natural
languages, as Chomsky called them, . . . [we]re languages in this sense.
The natural language English, for example, would be, in its written

98

Internalised language

form, a set of sentences each representable as a finite sequence of the

letters of the alphabet (Chomsky, 1957: 13).
Likewise, in its spoken . . . form, each sentence would be representable as a finite sequence of phonemes. Just as Im hungry would in writing be a sequence of orthographic characters, so the spoken sentence
corresponding to it may be represented by a sequence k + 0 + m + h + k
+ p + g + r + i, made up of the phonemes k, 0 and so on. In Bloomfields
terms, this would have represented a potential utterance: something that
a member of an English speech community could say. In Chomskys
terms, the natural language could be represented by a set of sentences
of which that is one member. But this formulation also made explicit
that some sequences of phonemes are not sentences. Thus the natural
language English would not be represented accurately, in its written
form, by a set of which one member is the sequence Am hungry I; or, in
its spoken form, by one which had a member a + m + h + k + p + g + r
+ i + k + 0. In Bloomfields terms, this would by implication not have
represented a potential utterance.
I have explained this with a care that is in part derived from hindsight.
At that stage, for example, Chomsky did not systematically distinguish
sentences and utterances. But if natural languages could be represented in this way, the fundamental aim in the linguistic analysis of a
language was, in his words, to separate the grammatical sequences which
were its sentences from the ungrammatical sequences that were not. It
was also, Chomsky added, to study the structure of the grammatical
sequences. But an essential task was to indicate what members the set
had. The grammar of a language will thus, as he defined it, be a device
that generates all of the grammatical sequences of [the language] and
none of the ungrammatical ones. So, for example, we can imagine a
grammar of English, as represented in its written form, that would generate a set with members such as Im hungry, You are thirsty, Are we
hungry?, Hes not thirsty; it would also say what structure each of these
has. But the set would not have among its members such sequences
as Am hungry I, Are you -y thirst?, or Nots angry he. Ideally it would
generate all the sequences, and only the sequences, that represent what
Bloomfield would have called the potential utterances of English.
This is from one page of Chomskys first book, and for most readers it
was a bombshell. For no one else had taken that to be the fundamental
aim of linguistic analysis. In the view then dominant in America it had
been precisely that of analysis: of the controlled division of utterances
into phonemes, morphemes, phrases, and so on. The overall description
of a language was a synthesis to which, as Harris put it at one point, the
work of analysis leads . . . up (Z. S. Harris, 1951: 372). A few years later

Generative grammars

99

Harris too was talking, as we have seen, of grammars generating sentences. But it was his pupil Chomsky who now took that goal as primary.
A generative grammar of a language such as English is essentially a
theory, as Chomsky put it in a later chapter, of that language (Chomsky,
1957: 49). Like any scientific theory, it was based on a finite number of
observations; and, in our case, these were observations of what people
say. A grammar of English was thus based on a finite corpus of utterances, taken to be in English. Any theory seeks to relate the observed
phenomena and, by constructing general laws, to predict new phenomena. Similarly, a grammar would contain grammatical rules which express relations among both the sentences of the corpus and the indefinite
number of sentences generated by the grammar beyond the corpus. These
were in turn its predictions. But how was the adequacy of such a theory
to be assessed?
One obvious test was whether the sentences predicted did count as
potential utterances. In Chomskys terms, we had to determine whether
or not [they] are actually grammatical, i.e., acceptable to the native speaker,
etc. (13). The etc. was important, and supplying what Chomsky called
a behavioral criterion for grammaticalness was never supposed to be
easy. But one external condition of adequacy (49) was that the sentences generated, those predicted as well as those observed, will have
to be acceptable to the native speaker. A grammar of English which
allowed a sentence like Am hungry I, or excluded Am I hungry?, would
accordingly be less adequate, in that respect at least, than one which did
not. Another external condition was illustrated in a later chapter with the
famous example of the shooting of the hunters. We saw earlier (in 5.3)
how this phrase was said to be understood ambiguously, and how, if a
grammar were simplified by the addition of transformations, its ambiguity
could be explained. In Chomskys terms, the transformational grammar
would treat it as a constructional homonym. Such homonyms existed,
by definition, whenever a given phoneme sequence is analyzed in more
than one way on a given level (86). Another criterion of adequacy, he
suggested, was whether or not each case of constructional homonym[y]
is a real case of ambiguity and each case of the proper kind of ambiguity
is actually a case of constructional homonym[y]. By this criterion, then,
a grammar of English which did not analyse the shooting of the hunters in
two distinct ways was less adequate, in that respect, than one which did.
A grammar was accordingly expected to predict the sentences or utterances of a language, and explain certain properties that they have. But
Chomsky also imposed a condition of generality. It was not sufficient
that a grammar of English or any other language should be externally
adequate. For that requirement might be met in many different ways.

100

Internalised language

Therefore, he argued, we must also prescribe the form a grammar may

take. Each grammar should, as Chomsky put it, be constructed in accordance with a specific theory of linguistic structure whose terms were
defined independently of any specific language (50).
In a footnote to this passage Chomsky referred to a paper published
three years earlier, in which Hockett had proposed criteria for models
of description. For Hockett too descriptions had to be such that, following the statements that they made, one could generate any number of
utterances in the language. They were thus predictive or, as Hockett
said, prescriptive; and the test for their predictions was, as he put it, that
of casual acceptance by a native speaker (Hockett, 1954: 232). A model
of description was defined by Hockett as a frame of reference within
which an analyst approaches . . . a language and states the results of his
investigations (210). This frame of reference would therefore have to
allow an analyst to formulate descriptions by which predictions were
made. It must, as Hockett put it, be productive: when applied to a given
language, the results must make possible the creation of an indefinite
number of valid new utterances (232).
The requirement that a description should be predictive was a criterion for descriptions; that a model of description should be productive was accordingly a metacriterion (232f.), and one of four that Hockett
had considered essential. The others were, first, that it must be general ;
secondly, that it must be specific; thirdly, that it must be inclusive.
To say that a model must be general was to require that it must be
applicable to any language, not just to languages of certain types; and
Chomsky, similarly, required that grammars should accord with a general
theory of linguistic structure. In arguing for this condition, Chomsky said
that, to assess the external adequacy of a grammar, we need to assume
no more than a partial knowledge of sentences and non-sentences. We
would know, for example, that Im hungry is an English sentence and Im
-y hung(e)r is not; but there would be many sequences of phonemes
about which we were not sure. Thus, to hazard an example, could one
or could one not say to someone Be hungry, you!? In many intermediate
cases we shall be prepared, he said, to let the grammar itself decide,
when it is set up in the simplest way so that it includes the clear sentences
and excludes the clear non-sentences. But [for] a single language, taken
in isolation, this was only a weak test . . . , since many different grammars
may handle the clear cases properly. To make it stronger he insisted that
they should be handled properly for each language by grammars all of
which are constructed by the same method (14).
To say that a model must be inclusive was to require that when [it is]
applied to a given language, the results must cover all the observed data

Generative grammars

101

and, by implication, at least a very high percentage of all the not-yetobserved data (again Hockett, 1954: 232). But we may make this more
precise if we think again of languages as sets to be generated. A model
could be seen as specifying, among other things, the form that, switching
now to Chomskys terminology, a generative grammar of a language
must have. A general model would have to be such that any natural
language can be generated.
To say that a model must be specific was to require that when [it is]
applied to a given language, the results must be determined wholly by the
nature of the model and the nature of the language, not at all by the
whim of the analyst. Hockett added that such a model might require us
to subsume . . . facts more than once, from different angles. But there
would be a lack of specificity if it allowed us to take our choice, instead
of forcing one or another solution. Now the purpose of Chomskys condition of generality was to reduce the multiplicity of grammars that, for
any language, might be externally adequate. Therefore a specific theory
of linguistic structure, as he described it, had to be specific in precisely
that sense.
In Chomskys early thinking, that translated, in particular, into the
requirement that a linguistic theory should provide an evaluation procedure (Chomsky, 1957: 51) for grammars. From what we have said already,
it was necessary to state precisely . . . the external criteria of adequacy.
This was in Chomskys account the first of three main tasks in the
programme he suggested (53). The second was to characterize the form
of grammars in a general and explicit way (54): to specify the kinds of
ways in which, as we have seen, they were to express relations among
sentences. But for any given corpus of utterances, there may still be
more than one alternative grammar, each in the prescribed form and
equivalent as to external adequacy, between which, if our theory was to
meet what Hockett called the metacriterion of specificity, it would have
to choose. The third task, therefore, was to analyse and define a notion
of, in Chomskys term, simplicity that we intend to use in choosing
among grammars all of which are of the proper form (54). The theory
would thus tell us, for any corpus and any two such grammars, which
is the better grammar of the language (51).
This was without doubt a tall order. A grammar was, once more, a
theory like, as a rave review proclaimed, a theory in the natural sciences
(Lees, 1957). But on top of that we then required a higher-order theory
of, in general, grammars. This not only specified the form that theories of
each individual language would have; it also had to show how one was
better than another. Such a theory would indeed have been a remarkable
achievement. For, as Chomsky himself remarked, there are few areas of

102

Internalised language

science in which one would seriously consider the possibility of developing a general, practical, method for choosing among several theories,
each compatible with the available data (53).
But it is easy to see why Chomsky did consider it. The science of
descriptive linguistics, as it was then called by Harris, Hockett and other
American scholars, had as its aim the description of languages, and a
grammar, in Chomskys sense, was such a description. But describing
a language is not like, for instance, describing a physical object. If I
describe a tree, say, or a picture, someone else might then in principle
examine it and judge directly how far my account is adequate. But to
describe a language is not so straightforward. We may describe, in something like the same sense, the behaviour of two people who on a particular occasion happen, among other things, to be making certain noises.
But the language they are speaking, however one defines it, does not lie
open to observation in the same way. For Saussure it had been a system
whose existence was not concrete. For Bloomfield in the 1920s it had
been a set of utterances; but, even if that definition is taken very literally,
we can never observe, and so describe in the ordinary sense, a language in
its entirety. So how is what we are calling a description to be validated?
The problem was faced by structural linguists from the beginning, and
the programme of controlled analysis, as envisaged by Chomskys American predecessors, had been one intended solution. It had led, in particular, to operational definitions of the phoneme (3.2), of the morpheme
(5.2), and increasingly, as Chomsky himself came onto the scene, of larger
units. Now for Chomsky a description was a grammar, and a grammar
was, more exactly, a theory. But his fundamental concern was still, he
said, that of the justification of grammars (49). A linguistic theory had
accordingly to be a higher-order theory, which in some way validated
object theories.
The second of Chomskys three tasks was, as we have seen, to characterize the form of grammars in general. But for this there was another
motive. For any linguist naturally had an interest both in the description of individual languages and in the properties that distinguish languages, in general, from other systems that are not languages. In Chomskys
terms, a particular natural language was the subject of a particular
grammar. This was, as we have seen, a theory of the language. The
properties of languages in general would in turn be characterised by
a higher-order theory satisfying his condition of generality. Now the
primary aim was, once more, to select for any language a particular
grammar from the multitude that would in principle be possible. But
the requirement was reasonable, Chomsky added, since we are interested not only in particular languages, but also in the general nature of
Language (14).

Generativeagrammars
Knowing
language

103

This last quotation is worth pondering. For structural linguistics had

been, from the beginning, a science of language systems. It was in essence
a linguistique de la langue; and implicitly, in any study of the wider
phenomena of langage, the findings of this discipline, concerning among
other things the character of langues in general, would be central. For
Chomsky it was similarly implicit that a theory of the general nature of
Language (singular and with a capital letter) would be, at its heart at
least, a theory of the general properties of (in the plural) languages.
6.2

Knowing a language

When Chomsky wrote Syntactic Structures he was still in his twenties;

and, however brilliant and original his ideas, the problems with which it
overtly dealt were still those of his older American contemporaries. But
by his own account his real goals were already changing.
A full exposition came in the 1960s, and began with a famous distinction between linguistic performance and linguistic competence. By
performance Chomsky meant the actual use of language in concrete
situations. So, if I come into the kitchen at a specific time and say Im
hungry, that is an instance of performance. But for me to say this, and
for whoever else is in the kitchen to understand me, we must share a
knowledge of the language I am speaking. This constitutes my or their
competence. Competence is thus the speakerhearers knowledge of his
language (Chomsky, 1965: 4); it is acquired in childhood by any normal
individual, and in Chomskys first conceptual shift, as he was later to
describe it, it became the central object of study.
There was a parallel, as Chomsky remarked, in Saussures theory of
langue and parole. Performance or parole was constituted by what
Bloomfield had called acts of speech: my act of speech on coming into
the kitchen, another act of speech by Bloomfields child who does not
want to go to bed, and so on. The study of a language as potential acts
of this kind had developed, as we have seen, into the proposal that a
grammar should generate utterances. But, for followers of Saussure,
linguistics in the strict sense was a theory of the nature and structure of
langues. In the Cours a langue existed in the minds of a community of
speakers, each of whom passively registers it (Saussure, 1972 [1916]: 30).
To describe it was to describe the sum of what could be so registered. For
Chomsky, competence was instead directly individual. But to describe
it was to describe the knowledge of a language that an ideal speaker
hearer, as an individual member of the community, would have.
For scholars educated in the tradition of Bloomfield this shift was
striking and controversial. But it is remarkable how smoothly ideas were
transferred from Chomskys earlier perspective to his new one.

104

Internalised language

In what, first of all, did competence consist? A grammar was, as we

have seen, a theory of a language. But to know a language native speakers
must, in Chomskys analysis, have formed analogous theories: they must
as children have worked out, from the speech they heard around them,
what are sentences that can be said in it and what structures they have. In
the 1950s, therefore, Chomsky had already suggested that, to have learned
a language, a child has in some sense constructed a grammar for himself
(Chomsky, 1959: 57). This was again a generative grammar; and obviously, as he now said, every speaker of a language has mastered and
internalized such a grammar (Chomsky, 1965: 8). Generative grammars
were thus, in the title of the section which includes this last quotation,
theories of linguistic competence. In a later passage, Chomsky suggested
that the term grammar could be used with systematic ambiguity. In
one sense, a grammar was the native speakers internally represented
theory of his language . This was what children were taken to construct
or master or internalise; and it was a system of rules that determine how
sentences are to be formed, used, and understood. The architecture
of this system was at the time the one described in our last chapter (5.3):
so, in constructing their theory, children had to work out the deep
structures that the sentences of their language have, how structures at
that level are related to surface structures, and so on. But in Chomskys
other sense a grammar was the linguists account of this system. So, in
developing as linguists our own theory of, for example, English, we
would be formulating an account (or grammar in the second sense) of
an internally represented theory (or grammar in the first sense) that
must be in the mind of anyone who knows English.
But if Chomskys theory of a language was now a theory of a speakers
competence, what of the higher-order theory, bearing on the general
nature of Language? This specified the form that grammars of a natural
language must take and a procedure that would justify a choice among
alternatives. But when children learn a language they were themselves
seen as constructing grammars. Those too would be theories based
on evidence of what people say: on observation, in Chomskys words,
of what we may call primary linguistic data. So, faced with this, a child
must have a method for devising an appropriate grammar. As a precondition for language learning, he must possess, according to the
sentence that immediately follows, both a linguistic theory that specifies
the form of the grammar of a possible human language and a strategy
for selecting a grammar of the appropriate form that is compatible with
the primary linguistic data (Chomsky, 1965: 25). A later section made
the parallel between children and descriptive linguists still more precise.
A child who is capable of language learning must have, Chomsky said:

Generativeagrammars
Knowing
language

105

a technique for representing input signals (compare a linguists phonetic

transcription of utterances); a way of representing structural information
about these signals (compare a linguists representation of, for example,
phrase structure); some initial delimitation of a class of possible hypotheses about language structure; a method of determining what each such
hypothesis implies with respect to each sentence; and, finally, a method
for selecting one of the ( presumably, infinitely many) hypotheses that are
allowed . . . and are compatible with the given primary linguistic data (30).
On that basis the term theory of language could in turn be used with
systematic ambiguity (25). Just as a grammar, or theory of a particular language, could be a system constructed in the minds of speakers, so
a theory of language in general would be in a corresponding sense what
speakers, when they are children, need in order to construct it. It was the
innate linguistic theory that provides the basis for language learning. But
in another sense it was a linguists account of what is innate, just as a
linguists grammar was an account of what has been learned. In this way
the whole programme of the 1950s, aimed at characterising languages
as sets of utterances, converted beautifully, if Chomskys reasoning was
accepted, into a programme for investigating both the knowledge that
each speaker of a language acquires and what Chomsky called the childs
innate predisposition to acquire it.
I have introduced these ideas as they were introduced by Chomsky
himself, and in the words he used in the 1960s. It will be noted, in particular, that the argument rested simply on the fact that people, in an ordinary sense, know languages. Knowing a language was then interpreted
as having a theory of a set of sentences; therefore people must, when
they are children, form such theories. A theory of a language was by
definition a generative grammar; therefore children must construct
or internalise generative grammars. They must be predisposed to do so;
therefore, it was argued, they must have foreknowledge of what languages in general are like, they must have a way of choosing between
grammars, and so on. So far, then, the reasoning was purely philosophical. It started from the initial statement that each language is known to
its speakers. From that one could not rationally dissent. The rest was an
analysis of what it means to have this knowledge and to come to have it.
But could there also be evidence that an innate linguistic theory exists? Let us assume that Chomsky, as a linguist, had formed a correct
account of what an ideal speakerhearer of, say, English must, as such a
speakerhearer, know. This knowledge has to be acquired in childhood:
therefore competence in English must develop in that form, from an
initial state in infancy in which a child has none (call it S0) to an eventual steady state (call it SS) in which, ideally, it is complete. How then do

106

Internalised language

the minds of learners get from S0 to SS? Now it might be that, if S0 were
blank, we could see no way by which, in principle, that could happen.
Therefore S0 could not be blank: we would have to ascribe to it specific
features that are needed if SS is to be reached. These features must be part
of just such an innate theory.
An innate grammatical theory was alternatively a Universal Grammar, and by the 1980s, in the period of what he called a second conceptual shift, Chomsky had long been convinced that there were
overwhelming arguments for it. In many cases that have been carefully
studied . . . it is a near certainty that fundamental properties of the grammars that children attain are radically underdetermined by evidence available to them, and must therefore be attributed to U[niversal] G[rammar]
itself (Chomsky, 1981: 3). This is cited from a technical monograph, in a
passage to which I will return. But the argument, as he had often put it
elsewhere, was from poverty of stimulus. The stimulus for acquiring a
language is formed by what he had called the primary linguistic data.
But the evidence from this is too limited for the knowledge that he
ascribed to adult speakers to have developed on that basis alone. Therefore, in his analysis, it must develop by an interaction of the input from a
childs environment with specific internal structures. These could only be
genetically inherited; or, in a loose sense, innate.
An early illustration was of structure-dependency in syntax. Take,
first of all, the sentence Is Mary coming? Its construction is that of an
interrogative; and in Chomskys early account of generative grammars,
just as passive structures were derived by a transformation from the corresponding actives (5.3), so the structure typical of questions was derived,
by an interrogative transformation, from that of statements. The structure
of Is Mary coming? was thus derived from that of Mary is coming. The
reason given was that, if we bring out in this way the parallel between
these structures, we can simplify the description of English syntax
(Chomsky, 1957: 65). But in Chomskys next phase learning a language
meant constructing a grammar. So, as the first step in his argument, we
assume that any speaker of English must construct one that includes this
transformation.
We also assume that the rule is as general as possible. It must therefore
cover not just sentences like this, in which the subject is the one word
Mary. In other cases subjects are whole sequences of words: thus, alongside Your sister is coming, with the subject your sister, there is a question
Is your sister coming?; alongside The girl next door cant come, where the
subject is the girl next door, there is a question Cant the girl next door
come?; and so on. A child must thus construct a grammar which includes
a transformation by which all such sequences are derived. But, in deriving

Generativeagrammars
Knowing
language

107

what is grammatical, the rule must also exclude what is ungrammatical.

Take, for example, the statement The girl who is invited is coming. Speakers
know that, alongside this, there is a grammatical sequence Is the girl
who is invited coming? Utterances like these may indeed be in the primary
linguistic data, for which a particular child internalising such a theory
must account. But adult speakers also know that, for example, one
cannot say Is the girl who invited is coming? The reason (we assume) is
that, when they were children, they internalised a grammar that includes
(we assume) a transformation, still constructed (we assume) in as general
a form as possible, which does not generate sequences of words of that
kind. But how, we then ask, were they able to do so?
In one seductive treatment Chomsky imagines a Martian scientist
(call him John M.) who is confronted with this problem. John M. would
discover a correspondence between statements and interrogatives: between
sentences like The girl is coming and The girl is invited and others like Is
the girl coming? and Is the girl invited? To be precise, he discovers . . . that
speakers form interrogatives by moving the verb to the front of the
sentence (Chomsky, 1988: 41). He also discovers that speakers can combine sentences like The girl is coming and The girl is invited to form more
complex structures like The girl who is invited is coming. How, he asks, do
they form an interrogative from that? Evidently, speakers have some
rule that they use to form interrogatives corresponding to declaratives;
this forms part of the language incorporated in their mind/brain; so his
task as a scientist is to construct a hypothesis as to what the rule is (42).
The obvious and most simple hypothesis, Chomsky supposes, is that
the rule obliges speakers to identify and move the first verb. Thus, in The
girl is coming, they find is and move it to form Is the girl coming? But if
that were indeed the rule then, from The girl who is invited is coming, they
would form precisely what they do not form: Is the girl who invited is
coming? Therefore the hypothesis fails. Is it possible then that they move
not the first verb but the last? This too is wrong: thus, to supply an
illustration, they do not move the last is in the statement She is the girl
who is invited to form an interrogative Is she is the girl who invited?
Eventually his Martian discovers that no rule will work if it refers simply
to the linear order of words. The rule that speakers have instructs them
to move neither the first verb nor the last; but what grammarians call the
main verb of the sentence (43). But to apply a rule in that form one must
undertake a complex computational analysis to identify the verb in question. One must analyse the structure of each statement, to identify the
subject phrase (the girl, she, the girl who is invited ) that the main verb
follows. Why, one may ask, do speakers use a computationally complex rule (44) instead of one which is much simpler?

108

Internalised language

John M. is puzzled, and might speculate that they are taught it. Perhaps, as children, they make mistakes and their parents or others correct
them. But he will quickly discover that this is not so. Children never
make errors about such matters and receive no corrections or instruction about them. Now the rule that Chomsky and his Martian have
proposed can be characterised as structure-dependent. Which verb is
moved depends on the grammatical structure that each sentence has. It is
thus a fact, as Chomsky puts it elsewhere, that without instruction or
direct evidence, children unerringly use computationally complex structuredependent rules and not ones that are computationally simple (Chomsky,
1986: 7). The mystery deepens: why do children not make errors and not
need instruction?
Chomskys Martian is left in the end with only one plausible conclusion. There has to be some innate principle of the mind/brain, which,
given the data that children have, yields this rule as the only possibility (Chomsky, 1988: 44). What is this principle? It is precisely that, in
general, rules of language . . . are structure dependent (45). They operate
on expressions that are assigned a certain structure in terms of a hierarchy of phrases: subject noun phrase, following verb, and so on. A child
learning any language knows, in advance of experience, that the rules
will be structure dependent. Therefore their rule for interrogatives cannot but be as the Martian, or as Chomsky, takes it to be.
Other arguments involved more complex evidence but had essentially
the same structure. Knowledge of a certain kind is attributed to state SS:
that of an ideal speakerhearer who has developed competence, as it was
called in the 1960s, in a language. We ask how SS can be reached; and
the only plausible conclusion is that, in an initial state S0, in advance
of experience bearing on its acquisition, certain general principles of
language are already genetically determined. By the 1970s, Chomsky was
already claiming that the growth of language could be seen as analogous to the development of a bodily organ (Chomsky, 1976 [1975]: 11).
Its development depended not just on the environment in which a child
was placed. It was not learned, therefore, in the sense in which we learn
things by instruction or by experience alone. It grew by the interaction
between input from experience and specific structures that were part of
our genetic inheritance.
If Chomsky was right this placed linguistics at the forefront of the
cognitive sciences. In the 1960s, when the idea was first put forward, it
caught the imagination of philosophers, psychologists, and others across
neighbouring disciplines, even though, at that stage, most of what he
later believed to be the structure of Universal Grammar, and therefore
what became his principal evidence for it, had been barely touched on.

Generativeagrammars
Knowing
language

109

Two decades later, he himself suggested that the discoveries he claimed

might be compared with the discovery of waves, particles, genes, valence,
and so on . . . in the physical sciences (Chomsky, 1988: 91). We are
beginning to see into the deeper hidden nature of the mind and to understand how it works, really for the first time in history. One hesitates to
cite a scholar when he is blowing his own trumpet, but if the discoveries
were genuine this was true. The main task for anyone who wants to
understand the nature of language would then be to extend them. It
was not to describe each language individually, each simply for its own
sake. It was not even to describe each individually, and then, by comparing them, to find out what, if anything, all have in common. It was to
study directly the genetically determined structures that, by Chomskys
argument, had to exist.
It was in the 1980s that the implications of this programme finally
became clear. There have since been changes of detail: many specific
hypotheses, in part modifying or overturning earlier hypotheses, about
the structure that the Universal Grammar has. But Chomskys basic
philosophy of linguistics has not altered.
To know a language was, he had said in the 1960s, to have constructed a grammar. Alternatively, we could talk of an internalized language, or I-language; so, as Chomsky now expressed what is essentially
the same point, a Universal Grammar could be construed as the theory of human I-languages. It was a system of conditions deriving from
the human biological endowment that identifies the I-languages that are
humanly accessible under normal conditions (Chomsky, 1986: 23). But
our problem then is to explain why individual languages, which all grow
in minds with the same biological endowment, are as different, even in
their basic structures, as they are. Let us suppose that, through our study
of an individual language, we discover that its speakers know x. This
might be, for example, a specific constraint on the movement of a unit
which belongs to a specific class. We are once more convinced that their
knowledge of x could not have been acquired without the mediation
of some universal principle. But suppose that x itself is found not, in
reality, to be universal. Thus, in some other languages, we find that units
of a similar class are subject not to x, but to another constraint y. Let us
suppose moreover that, for y to grow in speakers minds, we must again
invoke some universal principle. How can both x grow and y grow, if
the biological endowment of all speakers is the same?
The problem was in effect anticipated by the end of the 1970s. On the
one hand, Universal Grammar had to be inherited by every member of
our species; and, since any infant can be found in practice to be capable
of learning any language, there was no reason to suppose that it might

110

Internalised language

vary from one population to another. So, in Chomskys words, a theory

of it had to be compatible with the diversity of existing (indeed possible)
individual grammars (Chomsky, 1981: 3). On the other hand, it would
have to include whatever he considered necessary for the development of
any individual I-language. In Chomskys words, it had to be sufficiently
constrained and restrictive in the options it permits so as to account for
the fact that each of these grammars develops in the mind on the basis of
quite limited evidence. In the next sentence, which I cited earlier, Chomsky
said that it was a near certainty that many fundamental properties of
individual grammars had to be attributed to Universal Grammar. But by
then it was known that many of the arguments that justified that statement had been based on rules that varied from one language to another,
even within Europe.
The conclusion was that the inherited Universal Grammar had to include variables. By the basic argument, certain structures must originate
in S0; if not, Chomsky could not conceive how speakers minds attained
the steady state SS that he ascribed to them. The principles that determined these had then to be genetically inherited. But what is inherited
had now, in addition, to include explicit alternatives. The minds of
children about to develop a knowledge of English would still be initially
in the same S0 as those about to develop one of, say, Japanese. But the
task they faced was not just to acquire those aspects of I-language which
develop independently of Universal Grammar. Within the Universal
Grammar there would have to be specific choices, by which a child faced
with primary linguistic data in, say, English would develop an I-language
that is specifically of one type; but one faced with data in, say, Japanese
would develop another whose type might appear, at first sight, to be inexplicably different.
What we expect to find, in Chomskys own words, is a highly restricted
theory of U[niversal] G[rammar] based still on a number of fundamental principles that sharply restrict the class of attainable grammars
and narrowly constrain their form; but, in addition, with parameters
that have to be fixed by experience (again Chomsky, 1981: 3f.). If these
parameters, he continued, are embedded in a theory of U[niversal]
G[rammar] that is sufficiently rich in structure, then the languages that
are determined by fixing their values one way or another will appear to
be quite diverse, since the consequences of one set of choices may be very
different from the consequences of another set. Five years later, the
parameters of such a system were compared to an array of switches.
When a childs mind is in its initial state, each switch will be open; and, in
the transition from that to a final or steady state, each is set in either of
two ways. But for the core, as Chomsky called it, of each language that

Generativeagrammars
Knowing
language

111

is all that will be necessary. In the light of input from experience, a child
learning English will be led to make one series of choices. Each switch in
the array will then be set to correspond to one of two alternative values
of the relevant parameter. In the light of differing experiences, a child
learning Japanese will make choices that, in part at least, are different.
The corresponding switches will accordingly be set in ways that correspond, in part, to opposite values. But in either case no more will be
needed for the core of their I-language to develop. The transition from
the initial state S0 to the steady state SS was simply, in Chomskys words,
a matter of setting the switches (Chomsky, 1986: 146).
These passages are cited from the books that mark what Chomsky
himself saw as a new conceptual shift (again Chomsky, 1986: 6). It was
a shift internal to the theory of generative grammar; and, in the 1990s,
terminology from every phase of Chomskys exposition of his ideas
generative grammar, competence, I-language is still woven together
(Chomsky and Lasnik, 1995 [1993]: 1415). But the focus of this theory
had significantly changed.
In the beginning, as we have seen, a generative grammar was a set of
rules that generated, and assigned analyses to, a set of sentences. That
was how Chomsky had defined it in the 1950s (6.1), and it was such a
grammar that, in his first conceptual shift, was said to constitute the
competence of speakers of a language. In describing languages, a central
aim was thus to separate the sequences of words that were grammatical
from others that were ungrammatical. If we did not do this, we could
not claim that we had correctly characterised the competence, or knowledge, that their speakers had. But it was obvious in the 1980s that Ilanguages include much that cannot develop by the setting of parameters.
Speakers of English will know, for example, the forms of irregular verbs:
not thinked but thought, not haves sinked but has sunk, and so on. They
will know that boiler house or boat house are established combinations
with established meanings; but that bone house (in a poem by Seamus
Heaney) is not. They will know that there is a sentence It will be raining
cats and dogs tomorrow and that this means that it will rain very heavily;
they will know too that there is no corresponding passive Cats and dogs
will be being rained tomorrow. But details like these will not be determined
by the setting of parameters. They must be learned, from experience
alone, in just the way that, it was claimed, the core is not learned.
In Chomskys terms, such knowledge formed a periphery. The core
of an I-language was defined as a system determined by fixing values for
the parameters of U[niversal] G[rammar] (Chomsky, 1986: 147). This
was variously called core grammar or core language; and, in a later
account that was given in the 1990s, would consist of what we tentatively

112

Internalised language

assume to be pure instantiations of U[niversal] G[rammar] (Chomsky

and Lasnik, 1995 [1993]: 19f.). But the systems called languages in
common sense usage tolerate exceptions: irregular morphology, idioms,
and so forth (again Chomsky, 1986: 147). The I-language that a speaker
has developed will accordingly include not just the core, but also a
periphery consisting of whatever is added on in the system actually
represented in the mind/brain of a speakerhearer. In the 1990s this
periphery was again said to consist of exceptions (Chomsky and Lasnik,
1995 [1993]: 20).
The innate linguistic theory, as Chomsky called it in the 1960s, is
accordingly focussed on the core part of a speakers competence. But the
core is not a system that alone distinguishes what are sentences and what
are non-sentences. That will depend on both it and the periphery; and
exactly what belongs to which is not determined by our evidence of
any individual language that we might set out to study. What is an exception in the periphery is, presumably, whatever is not part of the
core. But what is part of the core can be determined only by a specific
theory of Universal Grammar, whose instantiation, on this hypothesis,
it will be.
The primary aim is therefore to engage directly with this aspect, as it
is claimed, of mans biological inheritance. But the grounds on which
the claim is made will also have to be more complex. In the beginning
evidence could come from a single language: if its speakers could be
shown to know things that they could not have learned from experience alone, a genetic explanation would be posited. This was made clear
by Chomsky in a footnote in the 1960s. Paradoxical as it may seem at
first glance, considerations internal to a single language may provide
significant support for the conclusion that some formal property should
be attributed not to the theory of the particular language in question
(its grammar) but rather to the general linguistic theory on which the
particular grammar is based (Chomsky, 1965: 209). I can recall one
reader who confessed to throwing down the book at this point. But, by
the logic of Chomskys argument, it had to be true.
We have seen how, in the 1980s, Chomsky could still argue in that
vein. But the theory of Principles and Parameters, as he calls it, is a
theory of everything in I-languages that is not an exception. Its parameters will therefore cover many things that can, in principle, be learned
from experience alone. One of Chomskys earliest parameters determined,
for example, whether subject pronouns could be dropped. English was
one language in which they could not: She is coming or He is coming,
not Is coming. In Italian the setting would be the opposite: Viene. But
there was no argument that, to acquire this one specific aspect of their

GenerativeGrammar
Universal
grammarsand diachrony

113

knowledge, children needed anything other than the evidence of speech

to which they were exposed. What was claimed instead was that this
difference was one of a series by which types of language were distinguished. In Italian, a verb like viene can be either preceded or followed by
a subject: Giovanni viene John is coming or Viene Giovanni. In English it
can only be preceded. This feature of Italian was then one of a clustering
of properties that, in Chomskys hypothesis, were related to this single
parameter (Chomsky, 1981: 240). Others were of a kind that, in the
1970s, had been taken to show that language could not be learned from
experience. But, once the parameter was set, through experience that
would be immediately decisive, they would necessarily follow. In English
the parameter would be set differently, and a different clustering of properties would follow likewise. As the values of this and other such parameters are fixed, the languages that are determined will, in Chomskys
words, appear to be quite diverse (4). If he is right, the explanation for
their diversity is that each parameter will have repercussions throughout
their core systems.
To justify the theory, it had now to be established that such clustering
of properties exists. But to that end it was not enough to look in detail
just at English or just at Italian. The validity of the theory depended on
the thorough study of the relevant structures in all languages, of every
possible type.
6.3

Universal Grammar and diachrony

Chomsky himself has said little or nothing about change in internalised

languages. But a theory has been developed by others, notably David
Lightfoot. His earliest treatment dates from the late 1970s, before the
second of Chomskys two conceptual shifts, and was inspired by accounts of Universal Grammar, or what was at first called an innate
linguistic theory, from the 1960s. His latest synthesis is, as I write, barely
a year old.
For Saussure, as we have seen, change in language was change in a
system. In one state a language has, for example, one system of consonants; in a later state it has another. Now a state of a language can,
for a follower of Chomsky, be at best a derivative concept. The primary
reality is the grammar constructed, in the terms he first used, by an
individual speakerhearer. To say that a language changes from one state
to another is accordingly to say that, in a later generation, children construct grammars differing in a certain respect from those constructed in
an earlier generation. Change in a Saussurean langue is at bottom change
in Chomskyan I-languages.

114

Internalised language

But for change in systems there must be a reason; and one suggested
by work on diachronic phonology (4.1) was that it might correct imbalance. Balance, by implication, was a property that systems tend to
maximise. Now the properties of grammars were the province of
Chomskys linguistic theory: both those that they must have and, since
he had talked in the beginning of a method for choosing between alternative grammars, those that are preferred. It was therefore natural, as
Lightfoot saw in the 1970s, that such a theory should be taken as a key
to understanding change. Suppose, in the extreme case, that a grammar
does not respect some limit that is laid down. It would then be driven, in
Lightfoots words, to a therapeutic restructuring (Lightfoot, 1979: 123).
He has tended not to refer to predecessors in the Saussurean tradition.
But, similarly, a theory of phonology might have defined the limits of
imbalance. Systems that breached them would be driven to change,
therapeutically, in such a way that they conformed.
In syntax Lightfoot proposed, in particular, that grammars were subject to a transparency principle. The simpler their account of the construction of sentences, and the closer the relation between surface structures
and underlying representations, the more transparency, as he defined
it, there would be. Grammars then tend to be restructured in ways that
will maximise it. But, as earlier linguists had seen, changes which are
therapeutic at one point in a system can create need for new therapy at
another. As Lightfoot put it, the restructuring of a grammar will be such
as to solve essentially local problems (378). But a local change will not
be inhibited if unwittingly it leads to problems elsewhere. [L]anguage
learners do not re-design their entire grammar or practise sufficient prudence to check all the implications of a given change for all other areas of
the grammar. Therefore a fresh problem may provoke another change.
This means, as Lightfoot said, that changes may often take place in
implicational sequences.
The assumption, to elaborate his image, was that grammars practise
therapy, not prophylaxis (123). But therapy itself could not be the whole
story. One is still left to wonder, Lightfoot said, why grammars never
reach a state of equilibrium or indeed what set off such implicational
sequences in the first place (381). Why do they not simply achieve a state
that is most highly valued, and stay put?
A similar problem had been implicit in the 1950s, and an answer, as we
have seen, was that the balance of a system, at one level, may be disturbed by cumulative shifts, at another level, in what Coseriu (4.2) called
a norm. Likewise, in Lightfoots terms, the causes of change were in
part extra-grammatical. Suppose, for example, that an ordering of words
falls out of use. Once it was normal; therefore instances of it were part of

GenerativeGrammar
Universal
grammarsand diachrony

115

the primary linguistic data to which children were exposed; therefore

the competence they acquired in turn permitted it. But, in performance,
other orderings were increasingly preferred: perhaps under the influence
of a foreign language (compare Lightfoot, 1979: 381f.), or for reasons of
expressivity (384f.). Such causes are beyond the scope of grammars. But
children of a new generation would then have increasingly little evidence
that the ordering was possible. Therefore they might construct rules that
did not permit it; that might lower the value of their grammars as regards
transparency; and, through other changes, limits laid down by linguistic
theory might no longer be respected.
The picture that Lightfoot presented was accordingly of piecemeal
changes, such as the exclusion of a specific arrangement of words, that
result in steady complication of a grammar, rendering it . . . less highly
valued (78). In his main illustration, words in English like can, may or
will were in the beginning verbs like any other. But, by what he described
as separate changes, they came not to take an object (one could not say
the equivalent of I can him or I may this); other verbs came to be followed
by an infinitive with to (modern I want to do it as opposed to I can do it),
and so on (details, 101ff.). Their status as verbs became less and less
transparent. What followed, in his account, was a restructuring of the
grammar in which they were no longer verbs, but formed a separate class
of modals. As such, they were subject to different rules. In Lightfoots
explanation, piecemeal changes can in this way lead up to a catastrophic
readjustment (78). As Coseriu would have said, before catastrophe theory
was invented, the system in time overturns.
These ideas had been inspired by Chomskys first conceptual shift;
and, since the object of investigation was not speech but a system underlying speech, it is natural that we should be able to find parallels in work
that looked back to Saussure. But Chomskys second shift soon followed,
and in Lightfoots response to that the model of therapy, in particular, is
transcended.
For a child constructing a grammar we must now talk of a child
setting parameters. To be precise, that is all that will be involved in the
construction of a Chomskyan core language. The primary linguistic
data that are available to children are accordingly a triggering experience (Lightfoot, 1999: 66f.) for Universal Grammar to develop into the
central systems of particular grammars. Changes in the basic structure
of a language then reduce to differences in the setting of parameters, in
response to different triggering experiences.
That much follows directly from what Chomsky himself proposed. But
setting parameters, as Lightfoot makes clear, is a task quite different
from the one that children had originally been said to face. According to

116

Internalised language

Chomsky in the 1960s, children did what linguists had been seen as doing
in the 1950s. In Lightfoots own terms, he viewed children as endowed
with a metric evaluating grammars which can generate the primary data
to which they are exposed, along with appropriate structural descriptions
for these data. This metric had to select the grammar which, among
other things, is most successful in generating those data (Lightfoot,
1999: 144).
In that account a child selects a grammar which matches his or her
input as closely as possible. But let us assume that the Universal Grammar is as Chomsky has claimed since the 1980s. In the development of
core structures, no set of rules will strictly be constructed: the possible
alternatives derive directly from the system that each child inherits,
and all that is going to happen, in the light of what Lightfoot calls the
triggering experience, is that one instantiation of a Universal Grammar
will be activated and the others suppressed. Nor does this require that
inputs should be matched. Parameters must be set easily, and the natural process is not one in which alternative core grammars, each derived
by different combinations of settings, are compared to see which, with
the addition of a periphery, generates most successfully the sentences
that a child hears. It will be one in which each setting is determined by a
minimum of specific evidence.
That insight is the key to everything that follows. In Lightfoots theory,
learning of a grammar, or core language, will be based on cues (149).
These are partial structures designated as such by a Universal Grammar;
and, in developing their knowledge of a language, children will scan
their linguistic environment (206 and elsewhere) for them. To be precise,
they will search for fragments of incoming speech that can be seen as
instances of them. Once a cue has been found sufficiently often or, as
Lightfoot puts it, with sufficient robustness the parameter that it represents will be set. But the Universal Grammar is a system, as we have
seen in the last section, in which the setting of a single parameter, which
will again be determined by experience, may have a wide range of consequences. Each of Lightfoots cues could therefore be associated with
a series of additional structures. Once the first choice is determined then,
without any further analysis of the input that a child receives, their developing grammar automatically includes these also.
For grammars to be different, there must first be a difference in the
triggering experience. This must again be due to nongrammatical factors (225), again involving, for example, language contact or socially
defined speech fashions (166). No theory of grammar or acquisition
(218) will predict shifts in the speech of a community at that level, and
Lightfoot says very little about them. But, given that they take place, he

GenerativeGrammar
Universal
grammarsand diachrony

117

now has an even more seductive theory of catastrophes. As speech shifts,

the evidence for specific cues may diminish. They may no longer be found
with sufficient robustness, and the parameters will no longer be set as
they have been. The expression of the cues will change, in Lightfoots
words, in such a way that a threshold [is] crossed and a new grammar [is]
acquired (166). But each cue is a structure that, once identified, entails
others. If it is not fixed, they are not selected either; and, as alternative
cues are identified, different structures will be selected, regardless of the
remainder of the input, by them. In this way, a small shift in the linguistic
experience of some children may have what Lightfoot calls dramatic consequences (111) for the grammars that develop.
It will be clear now why the notion of therapy was no longer needed.
We again speak, at the level of abstraction represented by the core of
Chomskyan I-language, of change in a system. To explain it, we must
again invoke shifts at another level, in the speech that supplies triggering
experiences. But the rest appeals simply to the laws of Universal Grammar, which inexorably determine specific structures in response to specific
inputs.

118

Structural semantics

Structural semantics

The term semantics was coined in the 1880s, by the French linguist
Michel Bral. As phontique was in its widest sense a science of the
sounds of speech, so smantique, from the Greek verb for to mean,
was a science des significations (Bral, 1911 [1897]: 8). How Bral saw
this discipline need not concern us. But for a structuralist a language is
a system in which determinate forms are related to meanings that are
also determinate. Structural semantics is accordingly concerned with
meanings as terms in such systems.
Its beginnings lie in Saussures theory of linguistic signs (2.1). Each
sign, as we have seen, had two sides; and just as on the level of expression, to use Hjelmslevs terminology, units were identified in abstraction
from the sounds that speakers physically utter, so those on the content
level had to be abstracted from the passing meanings that were intended.
This second task was therefore, in principle, as important as the first. But
when Hjelmslev wrote he was himself among the few who had approached
it seriously. We have seen how, by his test of commutation, both expressions and their contents were divided into figurae: [ram], for example,
into [r] plus [a] plus [m]; ram, in parallel, into he plus sheep (5.1).
Another parallel, as we will soon see, was between semantic oppositions
and distinctive phonetic features in Prague School phonology (3.2). But
these were ideas that few applications had explored in detail, and by the
1950s it seemed clear to many linguists that a great deal of the basic
structure of a language could be described, in practice, without doing so.
Structuralists in America, in particular, came to see the semantic system
of a language, as in Hocketts scheme of levels in the 1950s, as a field
which was separate from its central systems (5.2). Its investigation could
therefore, for the moment, be postponed.
For one contemporary commentator it was paradoxical and disturbing to find that many structural linguists are uninterested in problems
of meaning and reluctant to handle them (Ullmann, 1957 [1951]: 317).
It might likewise have disturbed Saussure, for whom each side of a linguistic sign had presupposed the other. But it was only after this stage
118

Meanings as invariants

119

had been reached that earlier ideas came finally to fruition. In the section
which follows the framework is still that of Saussure and of other structuralists whose ideas were formed before the Second World War. The
central problem, therefore, is the identification of what Saussure called
signs. In most later treatments the identity of units that have meaning
is one problem. The analysis of the meanings that they have can then be
treated as another.
7.1

Meanings as invariants

In the beginning, forms and meanings were indissociable. Ram, for example, is a word in English, with the meaning given in dictionaries as an
uncastrated adult male sheep. But its status as a word depends precisely
on its having a meaning. If rem, in contrast, is not a word in English, it
is because it has no meaning in English. That much was obvious; and
it was also obvious, and acknowledged since antiquity, that the forms
and meanings of words stood in a conventional or, in the term used since
the Renaissance, in an arbitrary relation. But what holds for words held
also, in Saussures analysis, for the meanings themselves. The meaning
uncastrated adult male sheep is arbitrarily related to ram. But its status
as a meaning depends precisely on there being a word whose meaning it
is. If, for example, no dictionary in English includes a meaning castrated
male dog, it is because there is no word in English for it. As the system
of the language distinguishes forms that are arbitrary, so it also distinguishes meanings that are arbitrary.
The underlying insight was that meanings too are different from one
language to another. Their speakers talk, in part, about the same world
and the same things and same happenings in it. But any translator knows
that what x means in one language often does not correspond exactly
to what y means in another. In Saussures analysis, the concept x has
its place in one system of oppositions; the concept y its own place in a
different system. Each system must be analysed, in a later structuralist
slogan, in its own terms. In the theory as developed by Hjelmslev, a
comparison of different systems will establish common areas of purport. But formal oppositions between x and other content-units of one
system structure such an area in one way. Oppositions between y and
other content-units in another system structure it differently.
A famous illustration is of words for colours. A speaker of English will
describe a certain range of colours as red; another as brown; another
as pink, and so on. But within each range there are perceptible differences: a reddish brown is objectively not like a greyish brown. Therefore
different languages can distinguish ranges that are not equivalent. What

120

Structural semantics

is brown, for example, in English is not always, in French, brun, and

what is indifferently black in English was in classical Latin variously
ater or niger. To underline the point that he was making, Hjelmslev set
four colour terms in Danish beside those that most nearly correspond in
Welsh. One range of colours is in Danish grn (green), and of these
some are in Welsh gwyrdd. But not all: others, closer to the range of
Danish blaa (blue), are instead glas. Colours that are blaa in Danish
are again glas, but so too are some that in Danish are graa (grey).
Others that are graa are then in Welsh not glas but llwyd; still others
that in Welsh are llwyd are in turn brun (brown). In Hjelmslevs
diagram (see fig. 6)

gwyrdd
grn
blaa

glas

graa
llwyd
brun
Figure 6

the colour spectrum is again a purport, which the oppositions within

Welsh and Danish structure, arbitrarily in Saussures sense, each in
their own way (Hjelmslev, 1943: 48f.).
The case for structural semantics rests on such facts. But the particular
conclusion that Saussure drew from them was that neither forms nor
meanings had an identity independent of a system in which they were
paired. One could not talk of words such as tree or horse independently
of either trif and hhfs, as mental representations of sounds, or tree
and horse, as mental representations of trees and horses (compare again
Saussure, 1972 [1916]: 97ff.). Nor could one talk of either independently
of the oppositions between each pair and all others. To identify words
was to distinguish simultaneously, within an individual language system,
what distinctively means (or signifies) and what is distinctively meant
(or signified).

Meanings as invariants

121

For Bloomfield too the essential first step was to identify recurrent
likenesses in form linked to recurrent likenesses in meaning. A speaker of
English might say, on a specific occasion, I planted a tree yesterday;
another, on another occasion, might say Those trees are magnificent.
But to identify their language we must abstract specific forms, like trif,
and specific meanings, like tree, that are linked consistently across
utterances. Now the form trif was not literally a recurring sound. It was
a phonetic form abstracted, as an invariant, from the widely varying
sounds that speakers actually produce. Nor was its meaning literally to
be identified with any of the varying objects that they may refer to. Our
first speaker might have planted, for example, an oak and, for it to be
planted, it would have had to be, as trees go, quite small. In Bloomfields
terms, both its species and its size would be features of the situation that
called forth the utterance that is produced (Bloomfield, 1935 [1933]:
139). The other speaker might instead be talking, in a very different
situation, of a grove of mature redwoods. But the features that concern
us are just the invariants that constitute the likeness. We have therefore
to discriminate, as Bloomfield put it, between non-distinctive features of
the situation, such as the species, size or state of a tree or trees referred
to, and the distinctive, or linguistic meaning (the semantic features) which
are common to all the situations that call forth the utterance (141) of, in
this example, the distinctive form trif. The problem for semantics, as for
phonology, was to identify such invariants.
A concept of a tree is one thing; features of the situations in which
tree is uttered are another. Nor, for that matter, is an acoustic image in
the minds of speakers the same thing as a unit physically identified in
sounds. But, whether one was a mentalist or a physicalist, the application of such theories could be less straightforward than it might seem in
this instance. Take, for example, the -ed at the end of planted. For grammarians it is an ending of a category that they describe as past tense;
this is also the tense of, among others, painted in I painted a tree yesterday, or decorated in I decorated a tree yesterday. But what establishes
this category as a sign? If we take these sentences alone the problem is
sufficiently simple. The ending 0d is constant, and in uttering it a speaker
will refer to what has happened earlier than the time of speaking. But a
past tense does not always have the ending 0d: compare, for example,
bought or came. Nor are forms of this category always different from
those of other categories. In I have planted a tree an identical ending 0d
is assigned instead to a category that is traditionally the past participle; in
I bet you 10, said in making the bet, the form of a present is identical to
that of the past tense in I bet you 10 last week. Nor does the past tense,
whether in 0d or otherwise, refer always to things that have happened in

122

Structural semantics

the past. Compare, for example, its use in If I planted a tree tomorrow.
The time reference is future, as it also is in If I plant a tree tomorrow. The
difference lies instead in the likelihood, as the speaker intends or perceives it, of the planting taking place. Nor is the past the only tense that
can refer to things that have already happened: compare, for example,
the narrative present in Then, last week, I plant a tree. If past tense is to
be identified by a direct relation between forms and their meaning, the
meaning cannot simply be past.
As conventional dictionaries list different senses of words, so conventional grammars list different uses of such categories. In our last
example, the present plant is said to have a historic use; in I plant a tree
every autumn its use is habitual; in I plant it tomorrow it is used to refer
to the future; and so on. To establish categories as terms in a sign system
we must therefore identify a meaning that is invariant across each range
of uses; by which present and past tense, for example, stand in a constant
opposition. But how abstract must such meanings be?
The problem was clear in the 1930s, and was addressed in particular
in two articles by Roman Jakobson. Both dealt with Russian, one with
categories of the verb, including tenses; the other, longer and more contentious, with the system of cases. For cases too, grammarians distinguish different uses or individual meanings (Einzelbedeutungen), and
there is again no simple relation between the forms that different words
take and the categories to which, despite this independent variation in
both forms and uses, they are nevertheless assigned (compare Jakobson,
1971 [1936]: 24f.). The same problem arises; and the solution, as Jakobson
saw it, was to establish a system of oppositions between general meanings
(Gesamtbedeutungen), by which each case is consistently distinguished
from all other cases. Individual meanings would then be explained by
reference to these general meanings, and the individual contexts in which
they are found.
As categories are opposed to categories so, in the phonology of the
Prague School, phonemes are opposed to phonemes. In English, for
example, m was a phoneme distinguished from b, among others, and
characterised in part by a phonetic feature nasal. Similarly, accusative
in Russian was a category to be distinguished from among others nominative, and characterised by a general meaning that had to be discovered.
But let us look further at the features by which phonemes are defined.
One features of m is, as we have said, nasal; b and p, which are also
articulated with the lips, can be contrasted as non-nasal. In the same
way nasal n, which is alveolar, is opposed to non-nasal d and t, and
nasal p (ng) to non-nasal g and k. What then is meant by the term
non-nasal? In Trubetzkoys account, it meant precisely that one set of

Meanings as invariants

123

phonemes could be characterised, in part, negatively. The three nasals

(m, n p) were characterised positively, by what he called a mark
(Merkmal) of nasality. Their place in the system was as consonants
produced specifically with air flowing through the nose. In opposition
to these, the non-nasals (b, p; d, t; g, k) were distinguished by the
absence of this mark. Their place in the system was accordingly as consonants that were not produced specifically in that way. Correlations, as
he called them, were defined by parallel oppositions of this kind, in which
the phonemes of one set are marked, and the others are in contrast
unmarked (Trubetzkoy, 1939: 77ff.).
This doctrine was new when Jakobson wrote the first of his articles,
and he at once seized on it as a clue for the analysis of meanings. Take,
for example, the distinction in word-meaning between duck, which can be
used of either females or males, and drake, which is used specifically
of males. The latter is the marked term, the mark being the specific
feature by which it is so used. The other is unmarked and in consequence it might be said to have two uses. If someone points to a female
mallard and asks if it is the drake, one might reply No, it is the duck: in
that case duck is used specifically of one sex. But if they point to a mixed
flock and say Look at those ducks, it is used of the species, or the kind
of bird, in general. Does it follow then that duck has two distinct meanings?
If we follow the logic of markedness it does not. Drake is the marked
term and is used specifically of male ducks. Duck is, in contrast, unmarked.
But that means simply that it is not specifically restricted in the way
drake is restricted. Therefore it can be used freely, both, in one set of circumstances, of ducks that are female and, in another, of ducks in general.
The argument is modelled on Jakobsons own treatment of a contrast
in Russian, between a word for a donkey and another specifically for a
female donkey (Jakobson, 1971 [1932]: 4). Let us therefore return, in this
light, to the distinction between past and present tense. In Jakobsons
account of Russian, a past tense makes clear that, specifically, the action
belongs to the past. But the opposing present constitutes instead a typical unmarked category. It does not indicate specifically that actions are
in present time; nor even that, in contrast, they are not in past time. It
simply does not indicate specifically what the past tense indicates. Therefore it has, inherently, no temporal restriction (an sich zeitlich unbestimmt
ist) ( [1932]: 8).
Let us suppose that the present is similarly unmarked in English.
Plant(s), for instance, does not in itself have a specific time reference.
Therefore, in I plant it tomorrow, it can combine with an expression that
refers to the future; and, in that way, is used, in one individual meaning,
of an event in the future. It can also combine with an expression such

124

Structural semantics

as every autumn; hence the habitual use, which is another individual

meaning, in I plant a tree every autumn. In I plant a tree on Tuesday it
might be used with either meaning: next Tuesday or, conceivably, every
Tuesday. But which it has depends entirely on the situation, to use
Bloomfields term, in which this is said. A present can again combine, in
narration, with an expression such as last week. Hence one could say
either Then, last week, I planted a tree, where planted is in the marked
tense; or, with the temporally unrestricted present, Then, last week, I plant
a tree. That would be the equivalent, in grammar, of referring to a male
of the family of anatinae as either a (marked) drake or an (unmarked)
duck. Similarly, in Jakobsons account of case in Russian, the accusative is marked in opposition to the nominative. In his initial formulation,
the accusative always indicates that some action is as it were directed
to the object referred to . . . (1971 [1936]: 31). Thus, in the Russian for
to write books, the word for books will be accusative, since it refers to
the objective of the writing. But a nominative, in contrast, indicates
neither the presence of a relationship to an action nor the absence of
such a relationship (32). In some uses, therefore, its meaning does involve
one: thus the word for books will be nominative in the passive Books
were written. In others it does not. Thus, as the term nominative
implies, it may simply name things: books (nominative), newspapers
(nominative), and so on.
Unmarked terms can thus have a wide range of uses. But what of those
that are marked? In English, as we have seen, the past tense is not always
used of events in past time. Compare again the conditional If I planted
a tree tomorrow . . . In Russian, an accusative can also be used with prepositions: onto the table (accusative). These uses are again in specific
contexts. Thus, in the example in English, it is relevant both that tomorrow
follows and that if precedes. But the general meanings of the marked past
and the marked accusative must allow for them.
It follows that, in grammatical systems, the marks may be very abstract.
In particular, they may themselves be negative. Thus, in an analysis of
tense in English, we might begin by talking of a reality that is subjectively before the speaker: this includes both things perceived and things
anticipated. Let us call this R. The meaning of the past tense might
accordingly be characterised as not R. In I planted a tree yesterday, the
planting is removed from this reality in one way, because the speaker
remembers the event objectively as something that has happened in the
past. But in If I planted a tree tomorrow . . . it is removed in another way,
in that it is not likely. Both uses of the marked term are then covered.
The unmarked present is in contrast, more precisely, not specifically
not R .

Meanings as invariants

125

This analysis of English is not Jakobsons; but in his own account

of Russian cases the level of abstraction was breathtaking. In the system
in general, the relation of the nominative to the accusative is also that of
the instrumental to the dative: in his terminology, this is a Bezugskorrelation or correlation, we may translate, of relationship. What is
indicated by, in general, the relationship cases (Bezugskasus) is the
non-independence of the object referred to (65): this implicitly subsumes
the non-independence, in to write books, of the books from the writing.
In the passage referred to Jakobson remarked that it is characteristic
of his oppositions that the marked term is in reality throughout of a
negative type. The nominative and instrumental, which are unmarked in
the relationship correlation, are then characterised as not specifically
indicating this non-independence. In opposition to the nominative and
the accusative, the instrumental and the dative are marked terms in an
edge correlation (Randkorrelation). As such, they indicate that the
noun occupies a peripheral position in the overall meaning-content of
the utterance (46). Peripheries, as Jakobson points out, imply centres:
nouns in the nominative and accusative are thus characterised in effect
as not specifically having a non-central position. As the nominative
and accusative are to the instrumental and the dative so the genitives are
to the locatives: locatives peripheral, genitives not specifically peripheral. These in turn form the marked terms in an extent correlation
(Umfangskorrelation). A locative indicates a locality within which something happens: for example, in Moscow. But such a locality is not represented to its full extent in the utterance (59). A locative is thus a marked
term in an extent correlation, as, he argues earlier (38), is a genitive. For
example, in Ivans (genitive) book, Ivan is represented only to the extent
that the book belongs to or is written by him. Now the genitive has, in an
atomic treatment, many different, often contradictory, meanings (37).
Only by this degree of abstraction, to a mark whose characterisation is
again essentially negative, can they be united.
This tour de force was for a long time unique. But if forms and meanings were direct terms in a sign relation, a category with varying signifiers,
as in our example from English, could be identified only by an invariant
meaning that is signified. The question of general meanings had to
form the natural foundation, as it was for Jakobson, of all study of
grammatical systems ( [1936]: 23). If each category has a multiple set
of varying meanings, the interrelation between the signs and the meaning
is, as he put it, lost and the questions of meaning are unjustly excluded
from the field of sign-theory (unregelmig aus dem Gebiete der
Zeichenlehre . . . ausgeschaltet). Semantics would be left without an
object (24).

126

Structural semantics

7.2

Semantic fields

For the British linguist John Lyons, writing in the 1960s, both phonology
and grammar were concerned with the expression-plane of language, one
with sounds and the other with words and their formal combinations
(Lyons, 1968: 54). Semantics was the study of the plane of content and
was therefore separate, as in Hocketts scheme of levels (5.2), from both.
By then the analysis of words was widely seen as prior to a description of
their meanings. But by then too that branch of description had at last
substantially progressed, not least through Lyonss own work.
A basic insight was that words are related within what were called
fields. The idea was originally from Germany, where it had been central,
in the early 1930s, to a famous analysis by Jost Trier of a change in
vocabulary in the mediaeval period. The terms involved were in the sense
area of knowledge (Sinnbezirk des Verstandes); and, around 1200, this
was found to be covered in German by the three words wsheit, kunst and
list. A century later it was found to be divided instead into wsheit, kunst
and wissen. But the difference did not lie in a simple replacement of list
by wissen. In the earlier period wsheit had, in brief, been all-embracing:
knowledge or understanding in all spheres. But by the later period it was
used specifically of wisdom in a spiritual or religious sense. In the earlier
period kunst applied to the knowledge and manners of a courtly society;
list to technical skill and knowledge, with no courtly or class connotation. But by the later period kunst did not have that connotation either.
What was thus affected, in Triers analysis, was a whole in which each
term had a changing place. We have seen how, in a structuralist account
of sound change, a system as a whole was said to react to pressure (4.1).
But Trier was among the first to turn such arguments against the doctrine
in Saussures Cours, by which diachrony was reduced to individual
changes. In this Trier was perhaps more truly structuralist. But the division as drawn by Saussure was described at one point at least as ein
Ausweg der Verzweiflung, or counsel of despair (Trier, 1938).
Triers theories are expounded lucidly by Ullmann (1957 [1951]: 156ff.),
and I confess that I have not gone back in detail to his main work. But in
a later article on linguistic fields, he referred to structures of this kind
as living linguistic realities. The vocabulary of a language as a whole
was articulated into partial wholes, which were in turn articulated into
the individual words that were their members (compare Trier, 1934: 430,
and Ullmanns commentary). The identity of some potential fields will
now be obvious. One is that of words for colour, which Hjelmslev was
to describe, as we have seen, as an area of purport. Another unites
words such as English drake and duck; ram, sheep and ewe; or duckling

Meanings fields
Semantic
as invariants

127

and lamb. Another well-defined field, which became a standard illustration in the 1950s, was of kinship: this is structured in English into mother,
father, daughter, and so on.
The structuralists who developed this idea did not need to conceive of
fields as quasi-animate entities. But clearly, if the vocabulary as a whole
was one big network of relations, there were networks within it that
could easily be treated separately. The task for structural semantics was
therefore to develop methods for doing so.
For most linguists the comparison with phonology was again seductive. Phonemes were opposed one to another by distinctive features of
sound. They formed series in which different phonemes were distinguished in parallel, and the more such parallels there were the more tightly
organised the system was. Meanings could also be said to enter, as we
have seen, into correlations. As drake, once more, was to duck so
gander was to goose. As ram or ewe were to sheep so stallion or
mare were to horse, and so on. Analogously they were opposed by
distinctive features of meaning. Drake, for example, could be characterised by a set of features of which one is specifically male: let us write
this, in a notation that has its origin in Jakobsons work on phonology in
the 1950s, as + Male. Duck again is not specifically male, or Male.
Mare may be characterised in part as + Female; horse as both
not specifically female ( Female) and not specifically male ( Male);
and so on. Not all semantic oppositions would enter into correlations.
But for some semantic fields the model did seem promising.
This view was largely to prevail. But the analogy rested on two different assumptions, and it is important to separate them. The first is
that we are dealing with a network, again, of relations. That words
could be related to other words was a familiar notion, and at least one
important relation has a name much older than structuralism. Nice, for
example, is an antonym of, or is opposite in meaning to, nasty. Good
is similarly an antonym of bad. The relation of drake to duck is in turn
like that of gander to goose, involving, one might say, a different kind
of oppositeness. But these are precisely relations in meaning between,
as I have represented them in italics, words. The second assumption
is that there are entities called meanings that are parallel to sound units.
If nice is an antonym of nasty, it is because a meaning nice is distinguished in just that respect from a meaning nasty. Our task once more
was to abstract and define these units, in terms of what they shared and
how they differed. This second assumption was long endemic among
linguists. But it does not follow from the first. Nor is it obvious that
relations among words, if we consider them as such, are like those between sounds.

128

Structural semantics

The study of relations as such was developed in Lyonss remarkable

first book, published in 1963. In later work they are described as sense
relations (thus Lyons, 1968: 428) and included some that were already
traditional, like antonymy or synonymy. But, as he points out, they were
generally treated in terms of a prior notion of meaning, independently
defined. For example, two words would be synonyms if their meanings,
seen as entities they designate, are the same (Lyons, 1963: 57f.). Likewise,
though the example is not his, the relation between nice and nasty would
be reduced to a prior difference between niceness and nastiness. Meaning was in this way seen as a more primitive notion (58), by which
semantic relations were in turn defined.
But in a structural approach, as Lyons interpreted it, meaning is
instead a function of these several relations. Let us take, for example,
the relation within the field of colour between red and green. If something
is red in colour it cannot be green, and vice versa. Likewise it cannot be
blue, yellow, and so on. In Lyonss terminology this is a relation of
incompatibility (59ff.). But to define it we do not need to establish in
advance criteria for redness, greenness or blueness. All we are saying
is that if, in certain sentences, the word red is replaced by green or blue,
what will be asserted cannot then, in the same circumstances, be true.
She wore a red dress will, as Lyons puts it, be understood as implicitly
denying She wore a green dress and She wore a blue dress (61). Another
relation within the field of colour is that of scarlet, for example, to red.
Lyons described this as a relation of hyponymy, or subordination in
meaning. But to establish it we do not first need to establish that a
meaning scarlet is included in a meaning red. We are simply saying
that the word scarlet and the word red will replace each other in sentences of which one unilaterally implies the other. Thus, as Lyons puts it,
X is scarlet will be understood (generally) to imply X is red ; but not the
reverse. Forms of antonymy are distinguished similarly (61ff.); also a
relation that Lyons calls converseness, between words such as husband
and wife. Thus, again, X is Ys husband implies Y is Xs wife; and, in this
case at least, vice versa (compare Lyonss definition, 72).
Part II of Lyonss book is a brilliant analysis of the terms for knowledge in the language of Platos dialogues. Three verbs can be translated
to know: in the infinitive, which Lyons uses as a citation form, epstasthai,
gign(skein, eidnai. Let me refer to these as know1, know2 and know3.
Associated with these are three nouns: tkhnB, which I will refer to correspondingly as knowledge1, gnsis (knowledge2) and epist*mB (knowledge3). What someone knows1 (epstasthai) is thus a knowledge1
(tkhnB), and so on. At the other extreme there are, for example, terms
for particular occupations: kerames potter or astronmos astronomer.

Meanings fields
Semantic
as invariants

129

Now the evidence from the texts is that such a person had a corresponding knowledge1: for example, a potter (kerames) knew1 (epstasthai) a
knowledge1 (tkhnB) that was specifically the knowledge1 pertaining to
pottery (keramik* tkhnB). Such a person was also a dBmiourgs, literally
someone working ((e)rg-) for a community (dBm-). It was thus part of
the meaning of this word that its hyponyms included kerames (potter),
astronmos (astronomer), and so on. It was again part of the meaning of
our knowledge1 (tkhnB), which was what someone knew1 (epstasthai),
that its hyponyms included that of a potter (keramea) or that of an
astronomer (astronoma) (compare Lyons, 140ff.).
One field, as Lyons describes it, is thus dominated by, or structured
under (140), words associated with know1 (epstasthai). It is again part
of the meaning of the subordinate terms, such as the Ancient Greek
words that we translate as astronomer and astronomy, that they belong to it. What, by contrast, one knows2 (gign(skein) was in general
different. For example, one knew2 a person: that is, gign(skein is found
in the texts, far more than epstasthai (know1), with an object such
as Socrates (accusative SDkrtB(n) ) (199f.). One could also know2,
for example, a fact or a name. But what one either knew1 (epstasthai)
or knew2 (gign(skein) one could equivalently, on the evidence of
other passages, know3. As one could know2 facts or names so too one
could know3 (eidnai) them (205ff., 219). Pottery or astronomy was
a knowledge1 (tkhnB ), or something that one knew1; but it was also
something that one knew3 (eidnai) and so a knowledge3 (epist*mB).
Lyonss hypothesis was thus, in its most general form, as follows. Firstly,
as know1 was to knowledge1 (epstasthai, tkhnB) so know2 was to
knowledge2 (gign(skein, gnsis) and know3 to knowledge3 (eidnai,
epist*mB). That too is argued from the evidence of the texts, though in
this summary I have taken the liberty of begging it. Know1 is used
differently from know2; and, he argues, the two verbs can contrast.
But know3 (eidnai) is neutral (223) between them. In that sense
the field of knowledge3 (epist*mB) subsumes those of knowledge1 and
knowledge2.
This is a bare gist, and more could be added. For example, someone
who is a dBmiourgs (worker for the community) could be specifically a
kheirotkhnBs, literally someone with a knowledge1 (tkhnB) involving
hands (kheir-); the meaning of this is partly given by its hyponyms,
which include kerames potter or iatrs doctor, and by its incompatibility with, for example, geourgs farmer (175f.). But the method is
throughout to clarify the meaning of words such as know1 or knowledge3, for which I have deliberately given no further gloss, by the relations that they contract, in appropriate contexts, with others. A structural

130

Structural semantics

analysis is an analysis of sense relations, in the context of specific sets

of sentences in which words can be used.
It was on this basis that, at one point in his introductory chapters,
Lyons assessed the analysis of meanings into features. Such analyses were
by then called componential: for example, the meaning of mother had a
component female or + Female, that of father a component male or
+ Male. The question, therefore, was whether it would help to factorise
a relation in meaning, between mother and father, aunt and uncle, and so
on, into parts of that kind. In cases like this it could have advantages,
where the components to be abstracted, such as sex, can then be described in a neutral metalanguage (80). This could be seen as the metalanguage, though Lyons does not refer to him at this point, of Hjelmslevs
purport. But in other cases it did not help. The word high is, in the
illustration that he uses, to the word low as the word good is to the word
bad or the word big is to the word small. This was as easy to factorise as
any other proportional equation: ax (high) is to bx (low) as ay (good )
is to by (bad ), and so on (79). But we are still no clearer as to what these
words apply to. The component y, for example, would be common to
both good and bad. But as a component of meaning it is no more easily
described in terms of reference than is the meaning of good and bad
themselves (80).
Lyonss conclusion was that linguistic units should not be multiplied
beyond necessity. But by the time his book was published such analyses
were becoming, one might almost say, rife.
The first paper in whose title the term componential analysis appeared
was by the American anthropologist Ward H. Goodenough, and belongs
with a group of studies that deal mainly, though not exclusively, with
words such as mother and father. What, as Goodenough puts it, do I
have to know about A and B in order to say, for example, that A is Bs
cousin? Clearly, he continues, people have certain criteria by which
this judgment is made. What the expression his cousin signifies is, he
says, this set of criteria (Goodenough, 1956: 195).
For Goodenough this illustrated very well the general problem of
determining what a linguistic form signifies. But such criteria are analogous, he at once says, to the acoustical criteria that distinguish
phonemes. These can be analysed, as he puts it later, as combinations of
percepts which we conventionally describe with reference to the manner
of their production in speech (197). For example, one set of percept
values relates to place of articulation. Similarly, the units signified by
forms consist of combinations of percepts and/or concepts, such as, in
the field of kinship, those of the sex of individuals and the generations
to which they belong. In an intervening passage Goodenough proposes

Meanings fields
Semantic
as invariants

131

even closer parallels. The task of a semantic analyst is to find the conceptual units out of which the meanings of linguistic utterances are built
(196). These conceptual units will be what is signified. What we are given
are the conventional symbols of speech which more or less stand for
these units. Thus we know, from a linguistic analysis as it was seen by
Goodenoughs American contemporaries, that there are expressions such
as cousin or his cousin. The task of a semantic analyst is accordingly like
that of a phonologist who might start from texts in a written language.
These would be given, and the problem would be to establish sequences
of phonemes to which they relate. The problem in semantics is to establish the conceptual units to which, analogously, expressions like his cousin
relate.
The analogy makes clear that the symbols and conceptual units are
established on distinct planes. To find the latter, we will then begin by
noting contexts in which an expression is used. For example, my cousin
may refer, on some occasions, to a person who is the speakers fathers
brothers son: in a notation used by anthropologists, FaBrSo. On
other occasions, it might refer instead to, for example, a mothers sisters
daughter (MoSiDa); and so on. For Goodenough such denotative types
were analogous to a phonologists allophones (3.2). Thus, if phonologists
were in fact to start from written forms in English, they would find, as
we saw, that written l and written ll correspond sometimes to clear [ l]
and sometimes to dark [n]. Their task would then be to determine, by a
test of what the Americans called complementary distribution (see again
3.2), whether these were variants of the same or of two different units.
Similarly, for Goodenough, the next step for a semantic analyst was to
examine the mutual arrangements and distributions of such denotative
types, and in that way to arrive at types that are significative (197).
Goodenoughs paper dealt in detail with the field of kinship on the
island of Truk in Micronesia, and it is in such anthropological studies that componential analysis was first developed seriously. But such
accounts were soon proposed by linguists also, both in America and in
continental Europe; and, if we may judge by the absence of references,
for the most part independently. A field in general would be defined by
the component or components that a set of meanings had exclusively in
common: for example, words for animals might all have meanings partly
characterised by the features animate (+ Animate) and not human
( Human). Within them smaller fields would have additional components: thus the basic meanings of sheep, ewe, lamb, and so on might, in
addition, include a component + Ovine. Opposite meanings would be
distinguished by contrasting features: + Male versus + Female, and so
on. Eventually the analysis will reach meanings that cannot be factorised.

132

Structural semantics

Thus lamb in a figurative sense might have a meaning characterised in

part by + Human; but beyond that it will have to be described, as in a
dictionary, as person who is meek, endearing, innocent, and so on.
Several versions are discussed in later work by Lyons (1977: 31735).
The most widely cited, however, was that of Jerrold J. Katz and Jerry A.
Fodor, which was part of a semantic theory grafted, in the same year as
Lyonss first book, onto Chomskys formulation of a generative grammar.
The problem, as they presented it, was to account for the ability of
speakers to interpret sentences in abstraction from the settings in which
they might be uttered. Speakers of English will agree, for example, that
The bill is large is ambiguous (Katz and Fodor, 1963: 174). It has at
least two readings, with the bill referring either to a demand for money
or to the beak of a bird. One ability, then, that speakers have is to
detect . . . ambiguities and characterize the content of each reading of a
sentence (175). In The bill is large but need not be paid they will understand that the bill must be a demand for money; this shows that a
speaker can disambiguate (175) one part of a sentence (the bill is large) in
terms of another. They will at once recognise that The paint is silent is
anomalous in some way, while The paint is wet and The paint is yellow
are not. Thus another facet of the semantic ability of the speaker is that
of detecting semantic anomalies. A semantic theory of English must
account for these and other abilities; and part of it, it was widely
acknowledged and certainly true, was a dictionary (181). Thus, in the
dictionary which would form part of a theory of English, bill must be
assigned at least two different sets of semantic features: perhaps, as in
printed dictionaries, by separate entries. The features assigned to wet
and yellow must combine readily with those of paint; those assigned to
silent must not. Pay must have features that are not compatible, in the
construction of The bill . . . need not be paid, with those that would make
up the meaning beak of a bird. In addition to the dictionary, a semantic
theory then had to include processes of combination. In The bill was paid,
they will yield a meaning for the sentence as a whole in which the features
of pay combine with those of bill only in the sense of a demand for
money. In The paint is silent, they will fail to combine those of paint and
silent; and so on.
This semantic theory of a language was presented as additional to a
generative grammar that, in the beginning, did not have a semantic component. Its lower bound (175) was fixed accordingly, by the structures
that that grammar would itself account for. As soon as grammars did
include semantic components, this became instead a boundary within
them, initially at a level of deep structure. The upper bound was then the
limit of the grammar itself: of the system of rules that, according to

Meanings interpretations
Semantic
as invariants

133

Chomsky, formed, or formed a theory of, a speakerhearers knowledge

of a language. For Katz and Fodor, a semantic theory was required to
account for readings of sentences in isolation (177), not ones determined
by settings. Likewise, as one part of a theory of linguistic competence,
the semantic component of a generative grammar had to assign semantic
interpretations to sentences, in implicit abstraction from performance.
Katz and Fodors readings in isolation were equivalent, it would seem,
to these interpretations.
The distinction between competence and performance was related, as
Chomsky put it, to that of langue and parole (see again Chomsky,
1965: 4). But, crucially, a speaker knew not just a dictionary. Semantic
interpretations were precisely of sentences; and there a tangle of new
problems lay.
7.3

Semantic interpretations

The Chomskyan revolution had begun, and it would have suited no one
to relate the ideas of Katz and Fodor to pre-revolutionary antecedents.
But there is a parallel, in part, with what Bloomfield had said thirty years
earlier.
Semantics, as Bloomfield defined it, was the second of two phases
in the description of a language. The first was phonology, and its task
was to establish both the phonemes of the language and the phonetic
forms, or combinations of phonemes, that occur in it. The task of
semantics was then to tell what meanings are attached to phonetic forms
(Bloomfield, 1935 [1933]: 138). A phonetic form which has a meaning
was a linguistic form, and this could be anything from a sentence, as the
largest unit that has meaning in the language (2.3), to a morpheme (5.2)
as the smallest. For each such form our task, as we have seen, was to
establish a distinctive or linguistic meaning, in abstraction from the
range of situations in which it is uttered (thus again Bloomfield, 141).
Bloomfield did not talk, and for him at least it would not have made
sense to talk, of meanings in isolation. But linguistic meanings had to
be distinguished from what he called the non-distinctive features of the
situation; or, one might equally say, the setting.
In defining semantics Bloomfield said that it is ordinarily divided
into two parts, grammar and lexicon (138). A lexicon was a stock of
morphemes (162) and the meaning of a morpheme is called, a few
paragraphs earlier, a sememe. In girls, for example, the plural ending is
a morpheme with the meaning more than one object (216); girl would be
another morpheme with the meaning we may once more represent by
girl. Grammar, in turn, was constituted by the meaningful arrangement

134

Structural semantics

of forms (163). Thus, in girls, the plural morpheme is related, in one

such arrangement, to girl; the meaning of the whole, by implication,
would be more than one girl. In the girls this word combines, in a larger
arrangement, with the. This is traditionally definite and, in Bloomfields
account, is one of a class of morphemes used in referring to identified
specimens (203). By implication, the whole phrase would have the meaning more than one identified girl. In You helped the girls, this phrase is
part of a larger unit helped the girls, forming a still larger unit with you.
The meaning of the sentence is accordingly a function of the meanings of
these parts and of the specific syntactic constructions (184) in which they
stand. The linguistic meaning of you is that of an entity or entities that a
speaker addresses, and you is related to what follows in what Bloomfield
calls an actoraction construction (172 and elsewhere). Helped the girls
has in turn an actiongoal construction. Thus, implicitly, the linguistic
meaning is that of an actor who is addressed ( you) having performed in
the past (-ed) an action of helping (help) whose goal was more than one
identifiable girl.
The statement of meanings was, for Bloomfield, the weak point of
language-study and would remain so, certainly under his physicalist conception of science, until human knowledge advances very far beyond its
present state (140). We therefore had first to identify forms, under a
fundamental assumption (2.2) which implied that, whether we can state it
or not, each linguistic form has a constant and specific meaning (145).
The morpheme girl is one such form, and in the girls there are formally
distinguishable ways in which it is arranged in relation to other morphemes. The next and much more difficult step was to identify linguistic
meanings, both lexical and grammatical. But it is instructive to recast
what Bloomfield said in terms that are more like those of a Saussurean
sign theory. Morphemes and their meanings would form simple lexical
signs. The smallest meaningful arrangements, with their meanings,
would form simple grammatical signs. Other units, from a word up to a
sentence, would be complex signs, whose forms and meanings could be
derived, in parallel, from the smaller signs, lexical and grammatical, that
they include.
For Katz and his associates the semantic interpretation of a sentence
was arrived at by a similar process of amalgamation (Katz and Fodor,
1963: 197ff.), in which meanings from the dictionary are combined in
a manner dictated by a description of its syntax (Katz and Postal, 1964:
12). The main difference, however, is that in Bloomfields case we can still
speak of a sign relation: gifl, for example, was a phonetic form associated directly with a constant meaning girl. But in Chomskys current
theory, the phonetic interpretations were determined by a mediating

Meanings interpretations
Semantic
as invariants

135

level of surface structure (5.3). Semantic interpretations were in turn

determined by a level of deep structure, and the reason was precisely that
the elements and relations which were distinguished on the surface did
not bear a one-to-one relation to them.
What distinctions among meanings should a grammar then make? The
years immediately before and after 1970 were a period of confusion in the
history of generative grammar, in which many new ideas were explored
with more enthusiasm than care. It is therefore very difficult to disentangle a specific answer to this question. But let us take for illustration a
distinction that has often been drawn by grammarians, between effected
and affected objects. The sentence I painted a picture will normally mean
that, by painting it, the speaker has brought a picture into existence.
The object a picture will thus refer to something that results from, or is
effected by, their actions. But it is also conceivable that the speaker has
applied paint to a picture that existed already: compare, for example,
I painted over your picture. A picture would then refer instead to something merely affected by what happens to it. Does the sentence then, in
isolation, have two readings? If so, the semantic component of Katz and
Fodors grammar had to assign both semantic interpretations to it. Or
are these readings merely distinguished by the settings in which it is
uttered?
This distinction is one of many discussed in the late 1960s, in an influential study by Charles J. Fillmore. In his terms an effected object
was assigned, at the level of deep structure, to a category factitive; an
affected object, as we will see in a moment, to a different category
objective. But further distinctions could also be drawn between subjects.
In an earlier paper Fillmore had begun by comparing, for example, the
janitor as subject in The janitor opened the door with the door as subject
in the simpler sentence The door opened. Though he does not say so, the
first sentence would fit perfectly the terms in which its construction would
have been described by Bloomfield: The janitor (actor) opened (action) the
door (goal). The janitor, in particular, refers precisely to someone who
performs an act of opening. But the door, in the second sentence, did not.
The semantically relevant condition common to these sentences was
therefore not between their subjects. For Fillmore it was instead shared
by the door, as subject of the second, and the door as object in the first. In
both, by implication, the door refers to something affected by an opening. He went on to compare the role of with this key, in The janitor
opened the door with this key, with that of this key as a subject, in This
key opened the door. Again no constant semantically relevant function
was shared by the subjects: keys do not, like janitors, perform acts of
opening. A key was instead an instrument, in the events referred to by

136

Structural semantics

both sentences, by which opening of a door is brought about. Therefore

the semantically relevant function of this key as a subject was said to be
shared by the instrumental prepositional phrase with this key.
On such evidence, Fillmore did not believe that subject and object
are among the syntactic functions to which semantic rules must be sensitive (1966: 21). The subject the janitor would instead derive, in a way
that is set out in detail in his later paper, from an underlying phrase
whose relation to the verb is agentive. This was the category, as Fillmore
defined it, of the typically animate perceived instigator of the action
identified by the verb (1968: 24). As subject or object, the door would
derive from a phrase whose relation was objective. This was for Fillmore
the semantically most neutral of his categories; but conceivably should
be limited to things which are affected by the action or state identified by
the verb (25). Keys in turn are instruments by which doors are opened;
therefore, both the subject in This key opened the door and the prepositional with this key would derive from an underlying instrumental. This
was the category of the inanimate force or object causally involved in
such an action or state (24).
What was the nature of such categories? In Chomskys system of grammar, a semantic rule was sensitive to deep structures. Therefore they
were identified at that level. But they were also said to form a conceptual system, which was the same for all languages. Thus, in Fillmores
definition, they comprised a set of universal, presumably innate, concepts,
identifying types of judgment human beings are capable of making about
the events that are going on around them (24). He therefore speaks, in
his final paragraph, of a semantically justified universal semantic theory
or, in inverted commas, a semantic deep structure , beside which the
deep structures that Chomsky described were likely to be artificial (88).
But, for many who read this, the conclusion was much simpler. On the
one hand there were surface structures, in which all subjects, for example,
had the same role. On the other there were representations of the meanings
of sentences, or semantic representations. But no valid level of deep syntax lay between them. Therefore concepts underlying surface structures
were directly categories of meaning.
The case against deep structures was first made, for quite different
reasons, by James D. McCawley: initially in a postscript to a paper that
appeared with Fillmores (McCawley, 1976 [1968]: 92ff.). But for those
who agreed with him, the surface structure of a sentence was again not
to be specified directly. In Chomskys account, a grammar had included,
first, a series of rules that characterised a set of deep structures; then a
series of transformations (5.3) that derived surface structures from them.
For adherents of generative semantics, as the movement initiated by

Meanings interpretations
Semantic
as invariants

137

McCawley became known, surface structures derived similarly, by a series

of transformations that might now be long and drastic, from representations of meanings. Therefore, in proposing a grammar, the first task was
to characterise, or generate, these. The obvious problem, which many
generative semanticists soon came to see as spurious, was how semantic
representations generated by a grammar, or what Katz and Fodor had
called readings in isolation, were to be identified apart from meanings
in general. The only constraint was that semantic representations could
be related, in some way, to surface structures. Wherever a sentence could
mean different things in different settings we could say, in principle, that
it was ambiguous, and, in these settings, it was disambiguated in one way
or another. Wherever a sentence was odd in a specific setting, we could
say that this was a semantic anomaly that speakers could detect and
that a grammar should explain; and so on.
The polemics surrounding this movement have been surveyed in a book
by Randy Harris (1993), and need not concern us. But by the mid-1970s
a clear reaction had set in, both in the work of Chomsky, whose main
interest, as always, was in syntax, and in that of theorists concerned
specifically with sentence meaning. Its terms were also in part different.
In particular, ideas from philosophy and logic have since then been at
least as influential as those from within linguistics.
The essential point for Chomsky was that grammars had to include
some aspects of meaning. Rules of semantic interpretation must indicate,
for instance, how much is negated by a word like not: for example, I
didnt leave because I was angry might mean either that the speaker, being
angry, did not leave, or that, although they did leave, that was not
the reason. Rules must also indicate the antecedent of a pronoun that
requires one, or what kind of relation a word like who bears to what verb
(compare Chomsky, 1976 [1975]: 104). A grammar would in that way
distinguish what he called the logical forms of sentences. For example,
the logical form of Who said Mary kissed him? is illustrated by an expression for which person x, x said Mary kissed him? (99). But it did not
follow that a grammar had to do more. As Chomsky came to see it,
it was reasonable to say that, with the characterising of such logical
forms, the theory of grammar, which he also called at this stage sentence
grammar, ends (104). For, beyond that point, it might well be impossible, as he put it in an earlier chapter, to distinguish sharply between
linguistic and nonlinguistic components of knowledge and belief . An
actual language might result instead only from the interaction of several mental faculties (43). The sentence grammar was the domain of
one such faculty, which was to become that of an internalised language,
or I-language, in the term that he introduced a decade later (6.2).

138

Structural semantics

Beyond it, logical forms would then be subject to further interpretation

by other semantic rules . . . , interacting with other cognitive structures,
giving fuller representations of meaning (105).
This formulation was new, and Chomskys view of grammar was at
the time quite tentative (43). But by the 1980s the syntactic component
of what was now his Universal Grammar simply mediated between
representations in phonetic form, assigned by one interpretive component, and representations in logical form , assigned by another
(Chomsky, 1981: 17). The properties of L[ogical] F[orm] were, he said,
to be determined empirically. The term was merely intended to suggest
no more that in fact, the representations at this level have some
of the properties of what is commonly called logical form from other
points of view. But in Chomskys view the representations P[honetic]
F[orm] and L[ogical] F[orm] could reasonably be supposed to stand at
the interface of grammatical competence, one mentally represented system,
and other systems. These in turn included the conceptual system, systems
of belief, of pragmatic competence, in a sense that will be clearer in a
moment, of speech production and analysis, and so on (18). This part of
Chomskys hypothesis was unchanged in the 1990s. Thus the language
faculty, in its widest sense, is assumed to have at least two components.
One is a cognitive system that stores information: this includes again the
core assumed to instantiate Universal Grammar. The other consists of
performance systems that access information and use it in various ways.
The cognitive system is then taken, in particular, to interact with two
external systems: the articulatoryperceptual system . . . and the conceptualintentional system. Phonetic Form and Logical Form are the
two interface levels (Chomsky and Lasnik, 1995 [1993]: 2).
If Universal Grammar is delimited in this way, many of the crucial
problems of semantics lie outside it. But they are problems nevertheless,
and in the last quarter century many have continued to address them. A
history of their ideas would take us well beyond the structuralist tradition
as I see it. But, from a structuralist viewpoint, the main problem was that
sentence meaning has no single definition.
Let us suppose, to take a simple illustration, that a husband and wife
are about to have dinner. The husband asks What are we drinking? The
wife replies I put some wine in the fridge; the husband fetches a bottle,
opens it, and so on. Now what the wife has, literally, said is simply
that she put some wine in the fridge. This is a statement that could, in
principle, be either true or false; and, in one account, the meaning of the
sentence that she utters, I put some wine in the fridge, would be given by
a statement of the conditions under which it would be true. It is also
tempting to compare this literal meaning with what Katz and Fodor

Meanings interpretations
Semantic
as invariants

139

called a reading in isolation, or with Bloomfields linguistic meaning. But

in uttering a sentence with this literal meaning the wife is understood,
and means to be understood, as answering the husbands question. They
are going, she suggests, to drink wine; and, since she is perhaps already
bringing in the food, he acts as he does. It seems then that we need two
different theories. One is a theory of the literal meanings of sentences.
This would be a theory of linguistic meanings: of the semantics of the
language that the couple are speaking. The other would be a theory of
how sentences, with such meanings, are understood in contexts in which
they are uttered. Since the 1970s, this has in common usage been not part
of semantics, but a complementary theory of pragmatics.
I have put terms in inverted commas since, although this illustration
may seem perfectly straightforward, the distinction between what is said
and what must be understood was a crucial difficulty. The sentence that
the wife is said to utter is itself an abstraction. What she means or
intends by uttering it is already another. In attempting to explain her
meaning we are, among other things, ascribing to the sentence itself a
meaning which is a still higher abstraction. The problem was whether we
had coherent criteria for it.
In the terms inherited from the 1960s, meanings were assigned to
sentences by a grammar, and a grammar, as Chomsky had defined it,
was an account of a speakers knowledge of a language (6.2). It was
therefore natural to define linguistic meaning, or meaning in isolation,
as a meaning assigned by theories of linguistic competence. Our wife
and husband share, among other things, a knowledge of English. The
rules that constitute this knowledge will assign an interpretation to the
sentence that she utters and, by definition, they belong to a theory of competence. Such a theory is then complemented by a theory of performance,
seen by Chomsky as the actual use of language in concrete situations
(thus again Chomsky, 1965: 4). By implication, this was concerned with
principles that did not reflect a knowledge of particular languages.
Semantics, therefore, might be seen as dealing only with linguistic competence; pragmatics only with performance.
A proposal like this was explicit, in the 1970s, in a first book by Ruth
Kempson (Kempson, 1975: 207ff.). When it was written Chomsky had
yet to publish his doctrine of logical forms, or to talk, as he soon did, of
pragmatic competence as distinct from internalised language. But it
was already unclear whether knowledge of a grammar was the only knowledge of a language that its speakers had to have acquired.
Kempson was also among the first to argue, as a linguist, that the
meaning of a statement should be identified with the conditions under
which it would be true. That view is now widespread, and semantics

140

Structural semantics

generally perceived as truth-based or, in the term that is now standard,

truth-conditional. But a speakers competence, or knowledge of a language, is not evidently so narrow. Someone, for example, who knows
French knows, among other things, when they should say tu and when
they should say vous. This is a matter, however, of context: of the
person to whom they are speaking. Were we saying, therefore, that this
aspect of their knowledge is a topic of pragmatics? If so, since pragmatics
and grammar are complementary, the latter does not cover every aspect
of linguistic competence. Or was it too a topic of semantics? If so, sentences with tu, like Tu dois sortir You (familiar singular) must go out,
have one semantic interpretation; corresponding sentences with vous, like
Vous devez sortir You (non-familiar) must go out, have another. But
suppose, say, that a husband, on a particular occasion, uses vous to his
wife. It is not obvious that the appropriateness or otherwise of what he
says reduces to conditions under which it would be either true or false.
This was one of many cases where, as Stephen Levinson was to put
it some years later, languages encode or grammaticalize features of the
context of utterance or speech event (Levinson, 1983: 54). Pragmatics, as
the study of how utterances are understood in contexts, would accordingly include both context-dependent aspects of language structure (like
the difference between familiar and non-familiar) and principles of
language usage and understanding that have nothing or little to do with
linguistic structure (9). But let us now return to what we called a literal
meaning. In simple cases this could be equated with an account of truthconditions. Thus, in our example, I have put some wine in the fridge is true
if the speaker has indeed put some wine in the fridge; and to know its
meaning as a sentence is to know that. In other cases it cannot. Thus
Vous devez sortir would not have the same literal meaning, even when
one person is addressed, as Tu dois sortir. Can we say, as we at first said,
that semantics is a theory of literal meanings?
Suppose, to take another illustration, that I ask a guest Would you
like a cup of coffee? This constitutes an offer; and, if my guest says Yes,
please, I am expected to make good on it. But the sentence I have
uttered has a construction like that of a question: Did you like the wine?
or Had she enjoyed the concert? It is therefore tempting to say that,
literally, it has the meaning of a question. This literal meaning would be
a function of the meanings of its individual elements (would, you, and
so on) and of the syntactic constructions, including what grammarians
call the interrogative construction, in which they stand. But why have I
uttered such a sentence? It cannot be because I simply want this information. Therefore, just as our husband understands, in context, that the
wine his wife had put in the fridge was what she was intending they

Meanings interpretations
Semantic
as invariants

141

should drink, so my guest understands, again in context, that if coffee is

wanted I will supply it.
In the early 1970s utterances like this were called indirect speech acts.
Thus, literally or directly, I am asking a question; but, indirectly, the
words I utter are an offer. We might then suggest that only the literal
meaning would belong to an account of my or my guests linguistic competence. That account is once again a grammar. The rest remains a matter of pragmatics: of understanding sentences, with prior sentence
meanings, in use.
But suppose that, in a similar context, I say Would you enjoy a cup of
coffee? This too would have the literal meaning of a question; and, if
pragmatics is not concerned with knowledge of a particular language,
there seems no reason why my guest should not understand my utterance
in the same way. But, of course, I would not dream of saying it. Would
you like . . . ? is a specific formula, in English, by which offers are made.
Would you enjoy . . . ?, despite the similar meanings of its parts, is not. If
a grammar was a theory of a speakers knowledge of a language, it
should surely distinguish them. If it did not, it was because pragmatics is
implicitly not just concerned, in Levinsons words, with principles . . . that
have nothing or little to do with linguistic structure.
Indirect speech acts were discussed by Levinson (1983: 263ff.), but not
from this angle. For it was soon taken for granted, often without explicit
discussion, that a distinction between competence and performance was
not simple. As a speaker of English, I know how to phrase an offer; also
how, for example, to make requests with varying degrees of diffidence or
politeness. I know that some words are used normally by, or in speaking
to, children (compare Levinson, 1983: 8); I also know the forms of speech
one might use in addressing, for example, a dog. Are these aspects of
performance, since they concern use in specific situations? Or are they
aspects of competence, since they involve linguistic knowledge that is
internalised by specific individuals? The terms and criteria available were
obscure and contradictory; and it was natural to restrict semantics to a
narrow path that did not stray into this quicksand. In particular, it was
natural to borrow from philosophers the apparently clear criterion of
truth-conditions.
These were problems of the 1970s, and later theories of meaning have
often bypassed them. But to do so is to acknowledge that distinctive or
linguistic meanings of sentences, as conceived in Bloomfields structuralist theory in the 1930s, were abstractions that later theories, of semantic
interpretations assigned to syntactic descriptions in abstraction from
settings, had failed to define univocally.

142

Structuralism in 2000

Structuralism in 2000

Is structural linguistics still a living movement, and if not when did it die?
The answers will depend on what is meant by structuralism; and, as
we found at the outset, that is not straightforward. But for many commentators its creative phase has long been over. Its heyday lasted from
the 1930s, when it was named, to the end of the 1950s; and, throughout
that period, linguistics was dominated by it. But structuralism in America is said to have been overturned by Chomsky, and by the 1970s his
hegemony was world-wide. Other programmes have emerged since, some
opposed to and some simply independent of his. But few have reasserted
views that he rejected. Therefore structuralism, as an active source of
ideas, is indeed dead.
This account should not be seen as an Aunt Sally. Most linguists do
not now describe themselves as structuralists, even if they work in fields
like phonology, or with units like the morpheme, that the movement
defined. When they refer to structuralist approaches to their subject,
they will mean ones that have been superseded since the end of the 1950s.
To follow these overtly would not, therefore, be a good career move. But
if linguistics is no longer officially structuralist, many linguists are still
strongly influenced by structuralist ideas. These concern, in particular,
the notion of a language as a system and the autonomy of linguistics as
defined by it.
The structuralist bible is, by long consent, Saussures Cours. In his
classic history of linguistics Robins summarised its contributions under
three heads (Robins, 1990 [1967]: 220f.); and, in my own presentation
(2.1), I have in essence followed him. Let us therefore recapitulate them,
and ask how far beliefs have changed since it was published.
Saussures first major contribution, following Robinss order, was his
separation of synchronic from historical linguistics. A language is once
more a system and change in it will involve transition from one state of
a system to a new state that results from it. To study the history of a
language was to study such transitions; therefore each successive state
had to be known. But to study a language as it is at one time was to study
142

Structuralism in 2000

143

it in abstraction from its history. How it arrived in that state did not
have to be known and, if known, was irrelevant. Saussures analysis did
not come strictly out of the blue. But it differed from the philosophy
of Hermann Paul and many other contemporaries; and must be seen,
in Robinss words, as a major factor in the development of descriptive
linguistic studies in the twentieth century.
Robins was writing in the 1960s, when, if structuralism was moribund,
most observers certainly did not know it. But since then this aspect of
Saussures thought has been rarely challenged. For Chomsky in particular,
a language is a structure in the minds of individual speakers, and linguistic theory is a theory of such structures. But an individual I-language
(6.2) is again a fact of a quite different order from a change between Ilanguages developed by successive speakers in successive periods. A theory
of I-languages explains how individuals develop, in particular, core languages. But a theory of what causes them to develop differently in different periods is separate and subsidiary. In his magisterial Introduction to
Theoretical Linguistics Lyons integrated Chomskys early programme with
that of Saussure, in a way that formed the thought of many who have
since been active in the subject (Lyons, 1968). But theoretical linguistics
was tacitly synchronic. In contrast to most of its predecessors, Lyonss
introduction did not deal at all with processes of change in language. The
logic that then lay behind this seemed impeccable, and its consequences
are still generally accepted. Theories of the structure of languages are one
thing, and descriptions of individual languages are informed by them. The
problems of historical linguistics are another; and, as presented in many
recent textbooks, it is just one of a cloud of specialities that surround the
centre of the discipline.
The second of Saussures main contributions was the abstraction of
langue from parole. As languages were to be studied in abstraction
from their history, so too the linguistic competence of the speaker, as
Robins puts it, was to be distinguished from the actual phenomena or
data of linguistics. Robinss wording may perhaps be thought to savour
more of Chomsky than of Saussure; but, in distinguishing competence
and performance, Chomsky himself remarked on the relatedness of their
ideas (again Chomsky, 1965: 4). The main difference in the status of a
language is that an individuals competence, as Chomsky defined it, would
for Saussure have been an imperfect knowledge of a system whose reality
was supra-individual. In Saussures terms, a language could exist completely only as a fait social, or social product, in a community. For
Chomsky, the reality is precisely the I-language, as he later called it,
that is developed by each speaker. But the object of investigation is again
not speech, but what is claimed to underlie speech.

144

Structuralism in 2000

It is easy to find earlier theorists who did not take that view. Language, in the opinion of the Italian philosopher Benedetto Croce, was
perpetual creation: language here translates il linguaggio. Each new
utterance is, precisely, new; and to search for the language which is its
model (la lingua modello) is accordingly to search for the immobility of
motion (Croce, 1990 [1902]: 188). These words are from the final chapter
of his treatise on aesthetics, in which he argues that the foundations of
that subject and of general linguistics are identical. If they appear to be
different, it is because we are influenced by grammars in the traditional
sense, which encourage the assumption, as Croce put it, that the reality
of language consists in isolated and combinable words, not in living
discourses (190). Croces ideas were formed at the end of the nineteenth
century, more than a decade before Saussures were published, and through
his continuing influence Italy was one country in which structural linguistics did not flourish. But even in English-speaking countries, where
it did, the object of investigation was not always as Saussure defined it.
For Bloomfield in the 1920s, a language was, once more, the totality
of utterances that can be made in a speech-community (2.2). For Zellig
Harris, who took this idea to an extreme, the structure of a language
was not independent, but a product of the scientific study of the patterns
that can be abstracted from the sounds of speech. The definition of a
language as a set of sentences was also, under his tutelage, the one first
given by Chomsky (6.1).
It does not follow that Bloomfield and his successors were not structuralists. It is perhaps significant that Hockett, among others, did define
a language as a system to be inferred from the data of speech: in his
terms, which may or may not be thought to betray behaviourist leanings, it was a set or system of habits (Hockett, 1958: 137, 141f.). But still
less does it follow that, by the 1960s, Chomsky had turned his back
on structuralism. It would be more reasonable to argue, with John E.
Joseph in a recent essay, that in defining a speakerhearers competence
as the primary object of study, Chomsky introduced structuralism into
American linguistics, more fully than any of his predecessors (Joseph,
1999: 26).
Robinss third point was that the Saussurean langue must be described
as a system of interrelated elements, not an aggregate of self-sufficient
entities. Its terms are to be defined relatively to each other, not absolutely
(1990 [1967]: 221). For Saussure, it was specifically a system of values,
in which neither what means nor what is meant has an existence independent of a network of oppositions in which they are paired. Linguistic
signs were thus established by a symmetrical relation between signifiers
and, in a brutal translation that linguists usually avoid, signifieds.

Structuralism in 2000

145

Saussures definition of a sign has been widely acclaimed, first by

linguists and especially, since the 1960s, by literary and other semioticians.
For Umberto Eco, it anticipated and determined all those that have
followed (Eco, 1975: 25); and in an introduction to Saussure by Jonathan
Culler, who is himself a literary theorist, the arbitrary nature of the sign
had pride of place (Culler, 1976: 19ff.). Within linguistics, we have seen
how Saussures theory was developed by Hjelmslev in the 1940s (5.1),
and by Martinet in his theory of monemes (5.2). It is therefore worth
remarking that, in the passage I am citing, Robins did not refer to it. It
was in just this period that a general science of signs, which Saussure
had called smiologie, was beginning to take shape, in France especially.
But the development of semiotics was not part of linguistics. By the 1960s
many linguists did not think of words, or other units of a language, as
signs in Saussures sense. Nor, in English-speaking countries, had most
strictly followed him.
We must be careful, at this point, not to be deceived by terms. In
French, the word signe ( [sio] ) is transparently related to the ordinary verb to mean ([sio]ifier). Therefore, if a language is a system of
meanings ([sio]ifications), it is natural to describe its elements as [sio]es.
But the ordinary verb in English is to mean; and, although to signify
(phonetically [s0gn0fk0] ) is established as a borrowing from Latin, its
relationship to sign ( [sk0n] ) is in etymology and spelling only. When
English-speaking linguists use the term sign it is therefore mainly, as in
this book, a reflection of French or other continental European usage.
But the difference has not only been in terminology. How, for example,
does one identify the morphemes in a word like waiting? For Bloomfield
they were again forms that, according to his fundamental assumption,
should recur in utterances in parallel with meanings (2.2). But for Zellig
Harris, in particular, the description of a language was based simply on
the analysis of formal patterns. These could be expected, in the end, to
cast light on their meanings. But no light would be cast if meanings were
appealed to at the outset. The evidence by which units were to be identified was simply of recurring combinations of sounds, like [0p], that could
replace other combinations of sounds, like the [0d] of waited, in the context of a set of other combinations of sounds, like [we0t], whose members
could replace one another in turn.
I have explored the history of this idea elsewhere (Matthews, 1993:
111ff.); for Harriss philosophy of linguistics I hope I may also refer
to a brief obituary (Matthews, 1999). It was specific to him; for most of
Bloomfields followers, morphemes were still not identifiable by formal
patterning alone. But by the 1950s they were no longer anything like
Saussures linguistic signs. They were units at an abstract mediating level,

146

Structuralism in 2000

not in itself of either sound or meaning. As such they were related, on the
one hand, to the combinations of phonemes by which they were realised.
These often varied from one allomorph to another. On the other hand,
they were related separately to meanings; and, in principle at least, these
too might vary. When Robins was writing, the theory of levels proposed
by Hockett and his contemporaries (5.2) was already giving way to
Chomskys scheme of deep and surface structures (5.3), in which forms
and meanings were related still more indirectly.
It does not follow, once more, that the American structuralists were
not structuralist. For what was common to all schools was the principle,
in Robinss words, that units are defined relatively to each other, not
absolutely. Phonemes, for example, were identified in opposition to other
phonemes. In Bloomfieldian terms, their realisations were in contrastive
not in complementary distribution (3.2). Morphemes likewise were to
be identified in opposition to other morphemes. We do not approach a
language with a check-list of sounds, like [l] or [d] or [t], that we expect
to find as phonemes in it. Nor do we approach it with a check-list of
conceptual categories, such as past time or plurality or the person
speaking, on which we try to map grammatical units.
In phonology, no competent investigator would describe a language
otherwise. That in itself has been one of the triumphs of structuralism. In
grammar, however, matters are less clear, and it may help to spell out the
principle further.
Let us take, for illustration, the word we in English. It is traditionally
first person plural and, by ancient definitions, forms are first person if
their reference includes the person speaking and plural if their referents are more than one. The words in double inverted commas may be
taken to refer to categories that reflect perceptions of the world of which
we are part. But it is easy to multiply distinctions. We might on occasion
refer to a speaker plus a single person who is spoken to (I and thou).
On another occasion it might refer to the speaker plus one other person
who is not being spoken to (I and another); to a speaker plus at least
two people who include one who is being spoken to (I and thou and
another, I and thou and others); and so on. One could therefore
imagine an account of English which would treat this one form as a
homonym with several different meanings. In Katz and Fodors terminology (7.2), We are leaving would have a set of readings in which the
reference of we includes a person spoken to, and another set in which it
does not. It would then be partly disambiguated in, for example, We are
leaving but you are not. It would have an intersecting set of readings in
which we refers to just two people, and another in which it refers to more
than two. On that dimension it would be disambiguated in, for example,

Structuralism in 2000

147

We two are leaving. We could easily continue: for example, a difference

between sexes would be disambiguated in We are leaving with our wives.
But that is not what English grammars do say. In other languages
there are indeed distinctions like these, between forms whose reference,
for example, includes or excludes the person spoken to, or between a
plural and a dual. But we in English is again undifferentiatedly first
person plural. In a structuralist account this term describes its place
in a specific system, formed by oppositions between we and I (plural
versus singular), between both we and I (first person) and you (second
person), and so on. In other languages the systems may be radically
different. Some will have no form that we could call first person plural.
They might, for example, have one which will indicate involvement of the
speaker: this would appear in a translation of both I am leaving and We
are leaving. The same form might be required in a translation of both
I saw them or We saw them and They saw me or They saw us. There is
therefore no distinction like that, as it is called in English, between forms
as subjects (I, we) and as objects (me, us). How the speaker is involved,
and who else is involved, could then be indicated by distinctions among
other categories.
Each different language had, for structuralists, to be described in its
own terms. Like other slogans, this can easily be twisted or misunderstood. But it meant above all that we do not look in every language for
conceptual categories, such as one of we-ness, which in fact reflect
distinctions in the languages that we ourselves know. Structuralism in
practice was a technique by which errors like that were avoided.
But did it follow that whole languages were systems o tout se tient?
Let us grant that we must be described specifically in opposition to I, you,
and others. You, for example, is then undifferentiatedly second person.
But there is already a complication, in that you does not always refer to
people spoken to. It also has an indefinite sense: for example, in a proverb like You cant make a silk purse out of a sows ear. In a Saussurean
account, these either reduce to just one general meaning (7.1) or have
separate meanings entering into separate oppositions. But is it necessary
to assume that languages are systems in Saussures sense, in which every
unit has a role defined entirely by the relations that it bears, direct or
indirect, to other units?
For Saussure la langue was a system of relations, including those that
the Cours calls syntagmatic. For Chomsky, in the 1950s, a grammar
was instead a system of rules assigning structures to sentences. But it was
again a system in which everything holds together: alter or delete one
rule, and either the whole is incoherent or the language generated will
be different. The generative grammar of the 1960s paired determinate

148

Structuralism in 2000

semantic interpretations of sentences, via determinate deep structures and

surface structures, with determinate phonetic representations (5.3). To
know a language was to know the rules of such a system, and this presupposed an independent mental faculty, of language, whose development
could be investigated separately from that of others. Linguistics in the
strict sense had been, in Saussures terms, an autonomous science of
la langue. It was now at heart a science of grammars, whose central aim
was to investigate their nature and what made their acquisition possible.
Generativism was in these respects in tune with earlier structuralism,
and, to recall a general judgment of Giulio Lepschy, an heir to it (Ch. 1;
Lepschy, 1982 [1970]: 37). But was this conception of a determinate language system, in its Saussurean or Chomskyan version, right?
It is not my business to debate the issue. But what has become clear,
since the 1930s and still more since the 1960s, is that, if a language is a
system like this, it will have a scope much narrower than that of any lay
conception of a language, or of what it means to know one.
The system had always been an abstraction. Thus, for the phonologists
of the Prague School, phonemes were defined by features that distinguished them from other phonemes. Other phonetic features were not
relevant to the structure of which these units were the elements (3.2).
But let us return, for instance, to the difference in English between a clear
and a dark l. It is not between phonemes; therefore, in describing the
system, we abstract away from it. But the variation is not random; nor is
it determined wholly by the structure of our vocal organs. It is a peculiarity of English, or of specific dialects within it. One solution was to ascribe
such patterns to a norm established at a lower level of abstraction (4.2).
As Coseriu pointed out, a language (lengua) was, in its ordinary sense,
not just the system; but the system plus the ways in which it is normally
realised (thus Coseriu, 1962 [1952]: 68). Norms are of their nature not
determinate.
There were also disagreements as to what precisely the abstraction
should include. A language system was a system of communication
through speech, and, as such, was seen as separate from other systems
governing the lives of speakers. But what properties of speech were
relevant to it? One obvious problem was that of intonation. A sentence
like Shes coming could be uttered, for example, with a fall in pitch (`)
centred on the last accented syllable: thus, with that syllable in small
capitals, Shes `ing. This might perhaps convey a sense of relief, if the
fall is emphasised. But the pitch could instead rise: Shes ing. This
might on occasion convey a speakers surprise. Were intonations like
these also, in Saussurean terms, the signifying elements of signs? For
Martinet, a language was a system of double articulation (5.2), and such

Structuralism in 2000

149

processes were then marginal to it (sur la marge de la double articulation). In particular, he saw differences in pitch as a continuous scale,
without definite units (Martinet, 1970 [1960]: 21f.). But, for Hockett and
other contemporaries in America, a language such as English had indeed
a definite series of pitch phonemes, just as it had definite sets of vowels
or consonants. As the latter realised morphemes such as she or come,
so pitch phonemes entered into the realisation of intonational morphemes
(Hockett, 1958: 34f., 125f.).
The criteria of abstraction were to that extent uncertain. But the issue
came to a head with Chomskys shift of focus in the 1960s. A grammar,
as he had defined it, was a system of rules that generated a language (6.1);
and both could again be justified as abstractions. But the grammar was
then said to characterise a speakerhearers knowledge of his language
(once more Chomsky, 1965: 4). It was therefore natural to ask if knowledge of these rules was all the knowledge of the language that its
speakers could be said to have. It seemed easy to think of other aspects
of, as Chomsky called it, their linguistic competence. If they know that
certain sentences are grammatical in it, they must also know a multitude
of norms of usage that are not so readily reduced to rules. If speakers of
English know that Is he coming? is a question, they must also know
that Would you like a cup of coffee? is a form of words appropriate to an
offer (7.3). They will know that, if they are trying to make an appointment, it would be normal to use a form like Can I see you tomorrow?
rather than, say, Can I come to you tomorrow? To speak the language as
others speak it they must certainly have mastered its intonations, whether
determinate or indeterminate. They must know which forms of speech,
intonations included, are polite and not polite; that certain forms are
usual only when one is in conversation with close friends; and so on. I
have put the word know in inverted commas, since what was called a
speakerhearers knowledge of his language was explicitly an idealisation. The implied criterion was that grammars were determinate and
existed independently of other mental systems. But this excluded much
that, in an ordinary sense, one could say about languages.
Within ten years it was clear, in particular, that it was seen as including only certain aspects of semantics. The semantic interpretation of
a sentence was, to be precise, its logical form; and, when Chomsky
introduced this term in the 1970s, he made clear that there were other
semantic rules, and other representations of meaning in a language,
beyond the scope of what he described at that time as a sentence grammar
(7.3). The apparent reason was that, if they were included, grammars
could no longer be seen as systems operating autonomously. The sentence grammar was the precursor of I-language in the 1980s; and, with

150

Structuralism in 2000

the theory of parameters, the system of core language was defined at

a yet higher level of abstraction. The scope of the periphery has at best
been illustrated; therefore it is not entirely evident what now characterises an I-language, with a core plus a periphery, as a whole. But the
core itself is a system that computes relations between Logical Forms
and Phonetic Forms. It was an abstraction within an abstraction, and
the criterion by which structures are included in it would be simply that
of relevance to Universal Grammar.
Chomskys theory of Universal Grammar was brand new, and addresses
problems that his structuralist predecessors had not dreamed of. It might
therefore be argued that the ideas of his second conceptual shift, as he
called it, are beyond my brief. But his is a very structuralist theory of
how knowledge of a language is acquired. The basic insight was that our
capacity for speech is, in part, genetically determined. When this was first
proposed it was a brilliant and exciting idea; and, in the light of what we
now know of communication in other species, and of other aspects of
behaviour in our own, it is at worst highly plausible. But it does not
follow that, within some larger mental structure, which in itself accounts
for only part of the way in which a community of people speak, there
is a further specific mental structure, supplemented and corrected by the
remainder, which is determined solely by a series of choices among
genetically determined options. This specific theory did not fully emerge
until the 1980s (6.2). But the first step was the reification of a generative
grammar, as Chomsky had conceived it in his first phase in the 1950s, as
(a theory of ) a speakers competence. The term grammar was defined
ambiguously, as we saw, with or without the words in brackets (again
Chomsky, 1965: 25). A linguistic theory, as again he had conceived it
in the 1950s, was then reified in turn, as a theory of specific mental
structures from which grammars develop. Ambiguously again, linguistic
theories simply were such structures.
The theory has also raised once more the question of how language
systems change. A language, in the formula in French so often said to
be Saussures, was a system in which everything holds together. But, if
that was true, why should it not persist in one fixed state? One possible
answer was that language systems are like thermostats or other servomechanisms, responding to external changes that affect them (4.1). Some
linguists have spoken figuratively as if, in this respect, a system had a will
of its own (4.3). Another possible answer was that even systems of this
kind had weak links or loose ends, and adjustments at these points could
lead to changes whose effects would then be wider. Another answer was
that languages were not only systems. Thus, for Coseriu, a shift in norms,
which were not determinate and changed gradually, could cause a system,

Structuralism in 2000

151

in turn, to overturn into a new state (4.2). For core languages to change
it seems that only such an explanation will be possible. The core is a
system in which everything, in the most striking fashion, holds together: change the setting of just one parameter, and the repercussions, it
is claimed, will be pervasive. How then can one generation come to set
parameters differently from earlier generations? The answer must be that
the language they hear, in a wider or more ordinary sense of language,
has changed independently. Explanations for that must again appeal to
shifts in norms, or to innovation under the influence of other languages.
Hence the diachronic syntax of Lightfoot (6.3) and others.
If languages have properties that change independently of grammars
or I-languages, it is natural to ask how such abstractions can be justified. The Chomskyan answer is obvious: that a Universal Grammar
exists as he says it exists. But what of linguists who are not Chomskyans?
When Robins was writing in the 1960s, the summary he gave of the
structural approach to language could be seen as underlying virtually
the whole of modern linguistics. It was in particular the foundation for
its autonomy (2.3), as a subject of study in its own right. Few linguists,
he added, would disclaim structural thinking in their work (again Robins,
1990 [1967]: 221). As I write, many more might now disclaim it. But for
how many is a language not, at some level of reality, a system that
defines their objectives?
The most obvious comment is that the discipline is no longer unified.
In the 1960s modern linguistics was still centred on what Saussure had
called la linguistique de la langue (2.1). The new theory of transformational grammar (5.3) was the latest attempt to clarify the structure of the
systems that were its object, and the range of facts that such a grammar
could capture, in a phrase then very fashionable, was compared directly
with the limited success, as it was claimed, of others. This led to intense
polemics between schools, whose aims, in developing a theory of a language, were implicitly similar. Other investigations into language, in
the wider sense of Saussures langage, were peripheral and were obliged
to take account of theories of that kind. In Chomskys terms, there
seemed little reason to question the traditional view that investigation
of performance will proceed only so far as understanding of underlying
competence permits (Chomsky, 1965: 10). It is remarkable how far his
contemporaries in fields such as psycholinguistics, both developmental
and experimental, have based their research on theories of linguistic structure that they took for granted.
But since the 1970s linguistics has fragmented, less into schools, as
in earlier decades, than into virtual subdisciplines, each with its own
objectives. Many are indeed concerned directly with langage, or with the

152

Structuralism in 2000

processes of speech (parole) as illustrated by a specific language. That

is true of much work on semantics and pragmatics since the early 1970s,
at least as I read it. Many studies are also truly interdisciplinary. The
scientific study of language, as some dictionaries define it, does not therefore still rest, in all branches, on the foundation that Saussure or
Bloomfield tried to give it. But many investigations do deal with specific
languages. How far does that part of the study of language remain structuralist at heart?
We must be careful, once more, to distinguish ideas that were specific
to structuralism. One is not a structuralist, for example, merely if one
uses the word structure. But the structure was of a language, and a
language was again an object of autonomous study. In a structuralist
view, the oppositions between I and you and we and you were part of the
structure of English, and the concepts that they represent internal to it.
The oppositions in French between je and tu or nous and vous were part
of another structure, which distinguishes its own concepts. We can subsequently compare such structures; we can then define a type of pronoun
system of which these are instances; we can also say that many other
languages have systems of the same type. We may then seek to explain
this by appealing to the world as we perceive it, in which speakers are
distinguished from people spoken to, in which one is different from
more than one, and so on. But from a structuralist viewpoint categories or concepts of specific languages were one thing. General categories
or concepts, which might be established independently or by comparisons between them, were another.
One foundation for this can be found in Hjelmslevs theory of form,
substance and purport (5.1). But there has been a reaction against structuralism in that form. One early instance was the assumption, in typologies
in the tradition of Joseph Greenberg, that in every language phrases can
be identified directly as the subjects and objects of sentences (4.3). But
when languages are considered in their own terms, categories like these
are not always evidently appropriate. Another was Fillmores proposal,
in the 1960s, by which sentences in any language would distinguish, at an
underlying level, a conceptual system of agentive, instrumental, and
so on. These were universals that concerned directly types of judgment
human beings are capable of making about entities participating in events
(7.3). Both suggestions could be taken to deny the structural autonomy
of individual languages; Fillmores, in particular, a distinction between
language and other mental capabilities. But these were matters on which
structuralism surely had to insist.
Thirty years have passed, in which the basis for proposals like these
has emerged more clearly. But one way to judge a discipline is by what is

Structuralism in 2000

153

usually taught to students; and, in the textbook introductions that most

lecturers recommend, the foundations of linguistics are still presented,
in the main, in terms in part structuralist and in part generativist. We
have seen already that on many issues, especially that of synchrony and
diachrony, these successive movements have been at one. A radical alternative, if indeed it has been truly formulated, has not yet won through.

154

References

References

Abercrombie, D. (1949) What is a letter?. Lingua 2. 5463.

Anderson, S. R. (1985) Phonology in the Twentieth Century: Theories of Rules and
Theories of Representations (Chicago: University of Chicago Press).
[Aronoff, M.] (1997) The editors department. Language 73. 4658.
Barr, K. et al., ed. (1938) Actes du quatrime Congrs International de Linguistes
(Copenhagen: Munksgaard).
Baudouin de Courtenay, J. (1895) Versuch einer Theorie phonetischer Alternationen:
ein Capitel aus der Psychophonetik (Strasbourg: Trbner). [Partial translation, omitting formalisms, in Baudouin de Courtenay, 1972: 144212.]
[Baudouin de Courtenay, J.] (1972) A Baudouin de Courtenay Anthology: the Beginnings of Structural Linguistics, trans. and ed. E. Stankiewicz (Bloomington:
Indiana University Press).
Bazell, C. E. (1953) Linguistic Form (Istanbul: Istanbul Press).
Bloomfield, L. (1914) An Introduction to the Study of Language (New York:
Holt).
(1926) A set of postulates for the science of language. Language 2. 15364.
[Reprinted in Bloomfield, 1970; Joos, 1958.]
(1935 [1933] ) Language (London: Allen & Unwin). [US edn 1933 (New York:
Holt).]
(1946) Algonquian. In H. Hoijer et al., Linguistic Structures of Native America,
85129 (New York: Viking Fund). [Reprinted in Bloomfield, 1970.]
[Bloomfield, L.] (1970) A Leonard Bloomfield Anthology, ed. C. F. Hockett
(Bloomington: Indiana University Press).
Boas, F. (n.d. [1911] ) Introduction to the Handbook of American Indian Languages
(Washington: Georgetown University Press). [1st edn in Handbook of American Indian Languages, Part 1 (Washington: Government Printing Office,
1911).]
Bral, M. (1911 [1897] ) Essai de smantique (science des significations), 5th edn
(Paris: Hachette). [1st edn, 1897. English translation, Semantics, by H. Cust
(London: Heinemann, 1900).]
Brndal, V. (1943) Essais de linguistique gnrale (Copenhagen: Munksgaard).
Brown, L., ed. (1993) The New Shorter Oxford English Dictionary, 2 vols.
(Oxford: Clarendon Press).
Chomsky, N. (1957) Syntactic Structures (The Hague: Mouton).
(1959) Review of B. F. Skinner, Verbal Behavior. Language 35. 2658.
(1965) Aspects of the Theory of Syntax (Cambridge, Mass.: MIT Press).
154

References

155

(1966) Topics in the Theory of Generative Grammar (The Hague: Mouton).

[Also in T. A. Sebeok, ed. Current Trends in Linguistic Theory, vol. I (The
Hague: Mouton, 1966).]
(1973) Language and Mind, enlarged edn (New York: Harcourt Brace
Jovanovich). [First published, without enlargements, 1968.]
(1976 [1975] ) Reflections on Language (London: Maurice Temple Smith). [First
published New York, 1975.]
(1981) Lectures on Government and Binding (Dordrecht: Foris).
(1986) Knowledge of Language: its Nature, Origin, and Use (New York: Praeger).
(1988) Language and Problems of Knowledge (Cambridge, Mass.: MIT Press).
Chomsky, N. and Halle, M. (1968) The Sound Pattern of English (New York:
Harper & Row).
Chomsky, N. and Lasnik, H. (1995 [1993] ) The theory of principles and parameters. In N. Chomsky, The Minimalist Program (Cambridge, Mass.: MIT
Press), 13127. [Revised version of paper first published in J. Jacobs et al.,
eds. Syntax: ein internationales Handbuch zeitgenssischer Forschung (Berlin:
de Gruyter, 1993).]
Coseriu, E. (1962) Teoria del lenguaje y lingstica general: cinco estudios (Madrid:
Gredos).
(1967) Larbitraire du signe: zur Sptgeschichte eines aristotelischen Begriffes.
Archiv fr das Studium der neueren Sprachen 204. 81112.
(1973 [1958] ) Sincrona, diacrona e historia: el problema del cambio lingstico,
2nd edn (Madrid: Gredos). [1st edn Montevideo.]
Cowan, J. M. (1987) The whimsical Bloomfield. Historiographia Linguistica 14.
2337.
Croce, B. (1990 [1902] ) Estetica como scienza dellespressione e linguistica generale,
ed. G. Galasso (Milan: Adelphi). [1st edn Palermo: Sandron, 1902.]
Crystal, D. (1997 [1987] ) The Cambridge Encyclopaedia of Language, 2nd edn
(Cambridge: Cambridge University Press). [1st edn 1987.]
Culler, J. (1976) Saussure (London: Fontana).
Delbrck, B. (1901) Grundfragen der Sprachforschung, mit Rcksicht auf W. Wundts
Sprachpsychologie errtert (Strasbourg: Trbner).
Eco, U. (1975) Trattato di semiotica generale (Milan: Bompiani).
Fillmore, C. J. (1966) A proposal concerning English prepositions. In F. J.
Dinneen, ed. Report of the Seventeenth Annual Round Table Meeting on Linguistics and Language Studies, 1933 (Washington: Georgetown University
Press).
(1968) The case for case. In E. Bach and R. T. Harms, eds. Universals in
Linguistic Theory, 188 (New York: Holt, Rinehart and Winston).
Firth, J. R. (1957) Papers in Linguistics 19341951 (London: Oxford University
Press).
Gabelentz, G. von der (1901 [1891] ) Die Sprachwissenschaft: ihre Aufgaben,
Methoden und bisherigen Ergebnisse. 2nd edn, ed. A von der Schulenberg
(Leipzig: Tauchnitz). [1st edn 1891.]
Gleason, H. A. (1961 [1955] ) An Introduction to Descriptive Linguistics, 2nd edn
(New York: Holt, Rinehart and Winston). [1st edn 1955.]
Goodenough, W. H. (1956) Componential analysis and the study of meaning.
Language 32. 195216.

156

References

Greenberg, J. H. (1963) Some universals of grammar with particular reference to

the order of meaningful elements. In J. H. Greenberg, ed. Universals of
Language, 73113 (Cambridge, Mass.: MIT Press).
Hamp, E. P., Householder, F. W. and Austerlitz, R., eds. (1966) Readings in
Linguistics II (Chicago: University of Chicago Press).
Hanks, P., ed. (1979) Collins English Dictionary (Glasgow: Harper Collins).
Harris, R. A. (1993) The Linguistics Wars (New York: Oxford University Press).
Harris, Z. S. (1941) Review of Trubetzkoy, 1939. Language 17. 3459.
(1946) From morpheme to utterance. Language 22. 16183. [Reprinted
in Joos, 1958; also in Z. S. Harris, Papers on Syntax (Dordrecht: Reidel,
1981).]
(1951) Methods in Structural Linguistics (Chicago: University of Chicago Press).
(1954) Transfer grammar. International Journal of American Linguistics 20.
25970.
(1991) A Theory of Language and Information (Oxford: Clarendon Press).
Hill, A. A. (1936) Phonetic and phonemic change. Language 12. 1522. [Reprinted in Joos, 1958.]
(1980) How many revolutions can a linguist live through? In B. H. Davis and
R. OCain, eds. First Person Singular, 6976 (Amsterdam: Benjamins).
Hjelmslev, L. (1938) Essai dune thorie des morphmes. In Barr et al., 1938:
14051. [Reprinted in Hjelmslev, 1959.]
(1943). Omkring Sprogteoriens Grundlggelse (Copenhagen: Munksgaard). [English translation, Prolegomena to a Theory of Language, by F. J. Whitfield
(Baltimore: Waverly Press, 1953; 2nd edn Madison: University of Wisconsin
Press, 1961).]
(1959). Essais linguistiques (= Travaux du Cercle Linguistique de Copenhague,
12) (Copenhagen: Nordisk Sprog- og Kulturforlag).
Hockett, C. F. (1942) A system of descriptive phonology. Language 18. 321.
[Reprinted in Joos, 1958.]
(1947) Problems of morphemic analysis. Language 23. 32143. [Reprinted in
Joos, 1958.]
(1954) Two models of grammatical description. Word 10. 21034. [Reprinted
in Joos, 1958.]
(1958) A Course in Modern Linguistics (New York: Macmillan).
Holtz, L. (1981) Donat et la tradition de lenseignement grammatical: tude et
dition critique (Paris: CNRS).
Jakobson, R. (1932) Zur Struktur des russischen Verbums. In Charisteria Guilelmo
Mathesio . . . oblata 7484 (Prague: Cercle Linguistique de Prague). [Cited
from Jakobson, 1971; also reprinted in Hamp et al, 1966.]
(1936) Beitrag zur allgemeinen Kasuslehre: Gesamtbedeutungen der russischen
Kasus. Travaux du Cercle Linguistique de Prague 6. 24088. [Cited from
Jakobson, 1971; also reprinted in Hamp et al., 1966.]
(1949 [1931] ). Principes de phonologie historique, trans. J. Cantineau. In
Troubetzkoy, 1949: 31536. [Reprinted in Jakobson, 1962; revised version of
Prinzipien der historischen Phonologie, Travaux du Cercle Linguistique de
Prague 4. 24767.]
(1962) Selected Writings, vol. I: Phonological Studies (The Hague: Mouton).
(1971) Selected Writings, vol. II: Word and Language (The Hague: Mouton).
Jakobson, R. and Halle, M. (1956) Fundamentals of Language (The Hague: Mouton).

References

157

Jespersen, O. (1922) Language: its Nature, Development and Origin (London:

Allen & Unwin).
Jones, D. (1962 [1950] ) The Phoneme: its Nature and Use, 2nd edn (Cambridge:
Heffer). [1st edn 1950.]
(1975 [1918] ) An Outline of English Phonetics, 9th edn (Cambridge: Cambridge
University Press). [1st edn 1918 (Cambridge: Heffer).]
Joos, M., ed. (1958) Readings in Linguistics, 2nd edn (New York: American
Council of Learned Societies).
Joseph, J. E. (1999) How structuralist was American structuralism? Henry
Sweet Society Bulletin 33. 238.
Katz, J. J. and Fodor, J. A. (1963) The structure of a semantic theory. Language
39. 170210.
Katz, J. J. and Postal, P. M. (1964) An Integrated Theory of Linguistic Descriptions
(Cambridge, Mass.: MIT Press).
Keil, H. ed. (18559) Prisciani Grammatici Caesariensis Institutionum Grammaticarum Libri XVIII, 2 vols. (Leipzig: Teubner).
Kempson, R. M. (1975) Presupposition and the Delimitation of Semantics (Cambridge: Cambridge University Press).
Koerner, K. (1989 [1987] ) Meillet, Saussure et la linguistique gnrale: une question dinfluence . In Practicing Linguistic Historiography, 40115 (Amsterdam: Benjamins).
Kurynowicz, J. (1949) La notion de lisomorphisme. In Recherches structurales,
4860. [Reprinted in Kurynowicz, Esquisses linguistiques (Wrocnaw/Krakw:
Polska Akadema Nauk, 1960).]
Labov, W. (1966) The Social Stratification of English in New York City (Washington: Center for Applied Linguistics).
(1977) Language in the Inner City: Studies in the Black English Vernacular
(Oxford: Blackwell).
Lees, R. B. (1957) Review of Chomsky, 1957. Language 33. 375 408.
Lehmann, W. P., ed. (1978) Syntactic Typology: Studies in the Phenomenology of
Language (Hassocks: Harvester Press).
Lepschy, G. C. (1982 [1970] ) A Survey of Structural Linguistics (London: Andr
Deutsch). [First published 1970 (London: Faber and Faber).]
ed. (1994) Storia della linguistica, vol. III (Bologna: il Mulino).
Levinson, S. C. (1983) Pragmatics (Cambridge: Cambridge University Press).
Lightfoot, D. W. (1979) Principles of Diachronic Syntax (Cambridge: Cambridge
University Press).
Lightfoot, D. W. (1999) The Development of Language (Oxford: Blackwell).
Lyons, J. (1963) Structural Semantics: an Analysis of Part of the Vocabulary of
Plato (Publications of the Philological Society, 20) (Oxford: Blackwell).
(1968) Introduction to Theoretical Linguistics (Cambridge: Cambridge University Press).
(1977) Semantics, 2 vols. (Cambridge: Cambridge University Press).
McCawley, J. D. (1976) Grammar and Meaning: Papers on Syntactic and Semantic
Topics (New York: Academic Press).
Martinet, A. (1949) La double articulation linguistique. In Recherches structurales,
307.
(1955) conomie des changements phontiques: trait de phonologie diachronique
(Berne: Francke).

158

References

(1970 [1960] ) lments de linguistique gnrale (Paris: Colin). [First published

1960. English translation, Principles of General Linguistics, by E. Palmer
(London: Faber and Faber, 1964).]
Matthews, P. H. (1993) Grammatical Theory in the United States from Bloomfield
to Chomsky (Cambridge: Cambridge University Press).
(1997) The Concise Oxford Dictionary of Linguistics (Oxford: Oxford University Press).
(1999) Zellig Sabbettai Harris. Language 75. 11219.
Meillet, A. (1937 [1912] ) Introduction ltude comparative des langues indoeuropennes. 8th edn (Paris: Hachette). [First edn 1912.]
Morpurgo Davies, A. E. (1998 [1994] ) Nineteenth-Century Linguistics. History
of Linguistics, ed. G. C. Lepschy. vol. IV (London: Longman). [Originally
La linguistica dellOttocento. In Lepschy, 1994: 11399.]
Murray, J. A. H. et al., ed. (1933 [18841928] ) The Oxford English Dictionary
(Oxford: Clarendon Press). [Reissue, originally in fascicles.]
Paul, H. (1920 [1880] ) Prinzipien der Sprachgeschichte. 5th edn (Halle: Niemeyer).
[1st edn 1880. English edn, Introduction to the Study of the History of Language, by H. A. Strong et al. (London, 1981).]
Percival, W. K. (1981) The Saussurean paradigm: fact or fantasy? Semiotica 36.
3349.
Piaget, J. (1968) Le structuralisme (Paris: Presses Universitaires de France).
Pos, H. J. (1939) Perspectives du structuralisme. Travaux du Cercle Linguistique
de Prague 8. 718.
Prague Linguistic Circle (1931) Projet de terminologie phonologique standardise.
Travaux du Cercle Linguistique de Prague 4. 30923.
Recherches structurales = Recherches structurales 1949: interventions dans le dbat
glossmatique (= Travaux du Cercle Linguistique de Copenhague, 5) (Copenhagen: Nordisk Sprog- og Kulturforlag).
Robins, R. H. (1990 [1967] ) A Short History of Linguistics, 3rd edn (London:
Longman). [1st edn 1967.]
Sapir, E. (n. d. [1921] ) Language: an Introduction to the Study of Speech (London:
Rupert Hart-Davis). [First published 1921 (New York: Harcourt Brace).]
[Sapir, E.] (1949) Selected Writings of Edward Sapir in Language, Culture,
and Personality, ed. D. G. Mandelbaum (Berkeley: University of California
Press).
Saussure, F. de (1879) Mmoire sur le systme primitif des voyelles dans les langues
indo-europennes (Leipzig: Teubner).
(1972 [1916] ) Cours de linguistique gnrale, published by C. Bally and A.
Sechehaye, with the assistance of A. Riedlinger. Edition with introduction
and notes by T. de Mauro (Paris: Payot). [1st edn 1916. English translation,
Course in General Linguistics, by R. Harris (London: Duckworth, 1983).]
Siertsema, B. (1965 [1955] ) A Study of Glossematics, 2nd edn (The Hague: Nijhoff ).
[1st edn 1955.]
[Sweet, H.] (1971) The Indispensable Foundation: a Selection from the Writings
of Henry Sweet, ed. E. J. A. Henderson (London: Oxford University Press).
Thibault, P. J. (1997) Re-reading Saussure (London: Routledge).
Trager, G. L. (1949) The Field of Linguistics (Norman, Okla.: Studies in Linguistics,
Occasional Papers 1).

References

159

Trier, J. (1934) Das sprachliche Feld. Eine Auseinandersetzung. Neue Jahrbcher

fr Wissenschaft und Jugendbildung 10. 42849.
(1938) ber die Erforschung des menschenkundlichen Wortschatzes. In Barr
et al., 1938: 927. [Reprinted in Hamp et al., 1966.]
Troubetzkoy, N. S. (1949) Principes de phonologie, trans. J. Cantineau (Paris:
Klincksieck).
Trubetzkoy, N. S. (1929) Zur allgemeinen Theorie der phonologischen
Vokalsysteme. Travaux du Cercle Linguistique de Prague 1. 3967. [Cited
from Vachek, 1964.]
(1931) Gedanken ber Morphonologie. Travaux du Cercle Linguistique de
Prague 4. 1603.
(1933) La phonologie actuelle. In Psychologie du langage, 22746 (= Journal
de Psychologie 30. 14) (Paris: Flix Alcan).
(1936) Die Aufhebung der phonologischen Gegenstze. Travaux du Cercle
Linguistique de Prague 6. 2945. [Cited from Vachek, 1964.]
(1939) Grundzge der Phonologie. (= Travaux du Cercle Linguistique de Prague
7). [French translation Troubetzkoy, 1949. English translation, Principles
of Phonology, by C. A. M. Baltaxe (Berkeley/Los Angeles: University of
California Press, 1969).]
(1958 [1935] ) Anleitung zu phonologischen Beschreibungen. 2nd edn (Gttingen:
Vandenhoeck & Ruprecht). [1st edn 1935 (Brno: Association Internationale
pour les tudes Phonologiques).]
Ullmann, S. (1957 [1951] ) The Principles of Semantics, 2nd edn (Glasgow: Jackson,
and Oxford: Blackwell). [1st edn 1951.]
Vachek, J., ed. (1964) A Prague School Reader in Linguistics (Bloomington: Indiana
University Press).
van Wijk, N. (1939) Phonologie: een Hoofdstuk uit de Structurele Taalwetenschap
(The Hague: Nijhoff ).
Vendryes, J. (1968 [1923] ) Le langage: introduction linguistique lhistoire (Paris:
Albin Michel). [First published 1923 (Paris: La Renaissance du Livre).]
Vennemann, T. (1975) An explanation of drift. In C. N. Li, ed. Word Order and
Word Order Change, 269305 ( [Austin]: University of Texas Press).
Wartburg, W. von (1969 [1943] ) Problmes et mthodes de la linguistique, trans.
P. Maillard. 2nd edn, rev. S. Ullmann (Paris: Presses Universitaires de France).
[Originally Einfhrung in Problematik und Methodik der Sprachwissenschaft
(Halle: Niemeyer, 1943).]
Weinreich, U. (1963 [1953] ) Languages in Contact: Findings and Problems (The
Hague: Mouton). [First published New York, 1953.]
Weinreich, U., Labov, W. and Herzog, M. I. (1968) Empirical foundations for
a theory of language change. In W. P. Lehmann and Y. Malkiel, eds. Directions for Historical Linguistics, 95195 (Austin: University of Texas Press).
Weiss, A. P. (1925) One set of postulates for a behavioristic psychology. Psychological Review 32. 837.
Whitney, W. D. (1867) Language and the Study of Language (New York: Scribner).
[ Whitney, W. D.] (1971) Whitney on Language: Selected Writings of William
Dwight Whitney, ed. M. Silverstein (Cambridge, Mass.: MIT Press).
Wundt, W. (191112 [1900] ) Vlkerpsychologie, vol. I: Die Sprache. 2 parts.
2nd edn (Leipzig: Engelmann). [1st edn 1900.]

160

Index

Index

Abercrombie, 34
acoustic images, 11f., 75, 93
affixes, 82
allomorphs, 86
allophones, 47
alternations, 32, 367, 39, 43
ambiguity
grammatical, 92, 99
semantic, 132, 137, 146f.
Anderson, 32
antonymy, 1278
arbitrariness, 17, 11920
articulation of language, 81, 83 4
associative relations, 50f.
balance, 57ff., 65, 114
Baudouin de Courtenay, 368, 403, 46
Bazell, 88f.
Bloomfield, 4, 11, 2132, 523, 82, 846,
968, 103, 124, 13345 passim, 152
on the autonomy of linguistics, 26ff.
his fundamental assumption, 23
on historical linguistics, 25f., 53
on linguistic meaning, 121, 1334
on morphemes, 82, 84, 133f.
as physicalist, 21f., 134
on the sentence, 28f.
Boas, 356, 39
Bral, 118
Brndal, 48
Brugmann, 8
change in language
compared to chess, 15, 53f., 61f.
in norms, 65f., 1501
phonological, 55ff.
as self-regulation, 58, 69
in syntax, 62, 703, 114ff.
as therapeutic, 55, 57, 73, 114
and universals, 70ff., 11317
and variation, 629

160

Chomsky, 3, 25, 68, 89117, 13250

passim
on competence and performance, 103,
151
his conceptual shifts, 96f., 103, 111
on deep and surface structure, 94f.
on evaluation procedures, 101f.
on generative grammars, 989, 1046
on I-languages, 109, 11112
on languages as sets of sentences,
989
on logical forms, 1378
on semantic interpretations, 925,
137f.
on transformations, 915, 99
on Universal Grammar, 106ff.
Chomskyan revolution, 25, 96, 133
commutation, 76ff., 81
competence, 1034, 14950
grammatical v. pragmatic, 138
v. performance, 103, 151
in semantics, 133, 13940
complementary distribution, 47, 856
componential analysis, 130ff.
components
of language system, see levels
of meaning, 131f.
concepts
in signs, 11f., 75
as universal, 136, 146, 152
constructional homonymy, 99
content v. expression, 7581
converseness, 128
core language, 11112, 116, 1501
Coseriu, 17, 636, 11415, 148, 150f.
Croce, 144
Crystal, 3
Culler, 145
deep structures, 945, 1356
semantic, 136

Index
Delbrck, 16, 28, 31
descriptive linguistics, 52
diachrony, Ch. 4, 11317
v. synchrony, 10, 1316, 25f., 52f.,
1423
distinctive features, 32, 35, 42f., 557,
122f.
Donatus, 81
drift, 71
duality, 84, 87
Durkheim, 12
Eco, 145
evaluation procedures, 101f.
expression v. content, 7581
features
of meaning, see semantic features
phonological, see distinctive features
figurae, 778
Fillmore, 1356, 152
Firth, 20f., 24
Fodor, 93, 1324, 137, 146
form v. substance, 789
Gabelentz, 6, 9, 26
gaps in systems, 5760
general meanings, 122ff.
generative grammars, 25, 90, 96ff.
change in, 11317
criteria for, 99102
and I-language, 109
and semantics, 92ff., 132ff.
as theories of a language, 99, 104
as theories of competence, 104
generative semantics, 1367
Gilliron, 54f., 57, 73
Gleason, 96
glossematics, 74, 81
Goodenough, 1301
grammar
generative, see generative grammars
v. lexicon, 133f.
v. phonology, 84ff.
grammatical v. ungrammatical, 98
Greenberg, 6970, 72, 152
Halle, 93
Harris, R., 137
Harris, Z. S., 245, 51, 90, 979, 1445
Heraclitus, 53
Hill, 45, 62f.
historical linguistics, see diachrony
Hjelmslev, 50f., 7481, 84, 86, 11820,
126, 130, 145, 152

161
Hockett, 45, 52, 83102 passim, 118, 126,
144, 146, 149
on the design of a language, 87
on models of description, 1001
on morphemes, 86f.
hyponymy, 1289
I-language, 109ff., 113, 1378, 14950;
see also generative grammars
idiolects, 66
indirect speech acts, 141
Indo-European, 710
innate linguistic theory, 105f.; see also
Universal Grammar
intonation, 1489
intrinsic meanings, 923
isomorphism, 80, 89
Jakobson, 40, 49, 558, 1225, 127
Jespersen, 53
Jones, 447
Joseph, 144
Katz, 93, 1324, 137, 146
Kempson, 139
Koerner, 6
Kurytowicz, 80, 89
Labov, 668
language change, see change in language
language system, see languages, langue
structure of, see levels
languages
as articulated, 81, 834
change in, see change in language
consistent, 70f.
as I-language, 109
as sets of utterances, 225, 978, 144
as social facts, 12, 63f., 143
structure of, see levels
as systems, 5ff., 9ff., 147ff.
as systems of values, 1620
langue, 1113, 27, 634
and competence, 103, 143
laryngals, 89
Lasnik, 111f., 138
Lees, 101
Lehmann, 701, 73
Lepschy, 34, 148
letters, 33f., 81f.
levels, Ch. 5 passim
of articulation, 83
of content and expression, 76ff.
deep and surface, 93f.
mediating, 86f., 935
of phonology and grammar, 84ff.

162

Index

Levinson, 1401
Lightfoot, 11317, 151
linguistic meaning, 121, 1334
linguistic signs, 1719, 74ff., 836, 118ff.,
134, 1445
linguistic universals, 70; see also Universal
Grammar
linguistics
as autonomous, 25ff.
descriptive, 52
historical, see diachrony
as science of languages, 5, 11ff., 103
logical forms, 1378
logical positivism, 21
Lyons, 126, 12830, 143
McCawley, 136f.
marked v. unmarked, 123f.
Martinet, 5862, 67, 73, 836, 88, 96,
148f.
meanings
as concepts, 11f.
of grammatical categories, 1225
as invariants, 95, 119f.
linguistic, in isolation, 93, 121, 1334,
139, 141
literal, 13841
of morphemes, 867, 1334
of sentences, see semantic interpretations
and transformations, 92
Meillet, 6f., 28
methods of analysis, 445, 4951, 85
models of description, 1001
monemes, 83; see also morphemes
morphemes
v. syllables, 82f.
as units of grammar, 8291
as units with meaning, 83, 867, 1334
morphophonemic system, 87, 92
morpho(pho)nology, 43
morphs, 86
Morpurgo Davies, 7f.
natural serialisation, 72f.
neogrammarians, 52
norm(s), 64ff., 114, 148, 150
operational definitions
of morpheme, 86
of phoneme, 47
Osthoff, 8
paradigmatic relations, 50f.
parameters, 11011, 11517, 151
parole, 11f., 103, 143

Paul, 26, 27, 52f., 143

Percival, 31
performance v. competence, 103, 133, 138,
143, 151
periphery v. core, 112f.
phonemes, Chs. 34 passim
change in, 55ff.
definitions of, 37, 43, 457, 93
as invariants, 44
v. morphemes, 82ff.
phones, 47
phonetic forms, representations, 935,
138
phonetics
anthropo- v. psycho-, 37
as peripheral system, 87
v. phonology, 402, 46
as study of speech sounds, 33
phonology
diachronic, 55ff., 114
v. grammar, 817
its influence in linguistics, 478
methods in, 44ff.
v. phonetics, 402, 46
phrase structure grammar, 90
physicalism, 21, 121, 134
Piaget, 69
pitch phonemes, 149
planes, see levels
Pos, 49
Postal, 134
poverty of stimulus, 106
pragmatics, 13941
Prague School, 40; see also Jakobson,
Trubetzkoy
primary linguistic data, 104
Principles and Parameters, 112f.
Priscian, 27, 82
psychophonetics, 37, 41, 46
purport, 80, 11920, 152
realisation
of morphemes, 85f.
as norm, 646
of phonemes, 44
relevance, 42f.
Robins, 1426, 151
Sapir, 26, 356, 3840, 43f., 46, 71
Saussure, 2, 4, Chs. 23 passim, 53, 746,
80, 93, 96, 103, 113, 11820, 126,
Ch. 8 passim
on associative relations, 50f.
on form v. substance, 789
and Indo-European, 89

Index
on langue v. parole, 1113, 27
on linguistic signs, 1719, 75, 11920
on the neogrammarians, 52
on semiology, 20
on synchrony v. diachrony, 10, 1316
smantmes, 82
semantic anomaly, 132, 137
semantic features, 121, 127, 130ff.
semantic fields, 126ff.
semantic interpretations, 925, 133ff.
as logical forms, 1378
semantics, 87, 925, Ch. 7, 149
in generative grammar, 92f., 132f.,
1368
v. phontique, 118
and phonology, 122f., 127, 1301, 133
and pragmatics, 13941
sememes, 133
semiology, 20, 145
sense relations, 128ff.
sentence grammar, 137
sentences, 259, 89f., 978
Siertsema, 75
sign relation, 75f.; see also linguistic signs
signifiant v. signifi, 16; see also
linguistic signs
significant differences, 345; see also
distinctive features
signs, sign theories, see linguistic signs
sound systems, 31ff.
change in, 55ff.
Stoics, 17, 81
structuralism
definitions of, 1ff.
origin of term, 48f.
structure-dependency, 1068
substance, 7881
substitution, 51
subsystems, see levels
surface structure, 95f., 1367
Sweet, 335, 39, 42f., 47f.
syllables, 813

163
synchronic v. diachronic linguistics, 10,
1316, 25f., 52f., 1423
syntactic descriptions, 935
syntagmatic relations, 27, 50f.
syntax, 89ff.
change in, 62, 703, 114ff.
system v. norm, 64f.
therapy, see change in language
Thibault, 11
Trager, 53
transformations, 90ff.
transparency principle, 114f.
Trier, 126
Trubetzkoy, 6f., 11, 409, 63, 74, 78, 92,
122f.
truth-conditions, 13941
Ullmann, 118
Universal Grammar, 10617, 138, 1501
argument for, 1068
and language change, 11317
universals of language, see linguistic
universals
unmarked, 123f.
utterances, 23f.
as generated by a grammar, 25, 978
Van Wijk, 48f., 55
variants, see realisation
variation, 629
Vendryes, 82
Vennemann, 703
Wartburg, 69, 73
Weinreich, 67f.
Weiss, 21f.
Whitney, 17
words
in antiquity, 81f.
v. morphemes, monemes, 82f., 88
Wundt, 268